Blood Vessels and Perivascular Phagocytes of Prefrontal White and Gray Matter in Suicide

Abstract

Inflammatory processes may contribute to psychiatric disorders and suicide. Earlier, we reported greater densities of perivascular phagocytes in dorsal prefrontal white matter (DPFWM) in suicide than in non-suicide deaths. To distinguish between greater vascularity and greater coverage of vessels by perivascular phagocytes, and to determine whether the excess of perivascular phagocytes is derived from microglia or from non-parenchymal immune cells, we made stereological estimates of vascular surface area density (A_VTOTAL) by staining for glucose transporter Glut-1, and the fraction of vascular surface area (AF) immunoreactive (IR) for CD163 (CD163 AF) in dorsal and ventral prefrontal white and gray matter. Manner of death or psychiatric diagnosis showed no association with CD163 AF in any region. Suicide was associated with a lower A_VTOTAL compared with non-suicides in DPFWM (p = 0.018) but not with A_VTOTAL in the 3 other regions of interest. Thus, the earlier observation of increased density of perivascular phagocytes in DPFWM after suicide cannot be attributed to infiltration by peripheral monocytes or to increased vascularity. Greater A_VTOTAL ventrally than dorsally (p = 0.002) was unique to suicide and white matter.

INTRODUCTION

Over 800 000 deaths by suicide occur in the world each year (1), and despite increasing pharmacological treatment and clinical vigilance, it is estimated that worldwide, 1.5 million people will die by suicide in 2020 (2). Thus, it is important to investigate clinical and biological triggers, mediators, and effectors of suicidal behavior. Ninety percent of people who die by suicide meet The Diagnostic and Statistical Manual of Mental Disorders (DSM) or The International Classification of Diseases (ICD) criteria for a psychiatric disorder, most commonly depression or schizophrenia (2). Medications targeting serotonergic neurotransmission have been the front-line pharmacological treatment options for depressed suicidal patients over several decades. However, their limited efficacy and a possibly increased risk of suicidal behavior at the onset of treatment (3) have impelled researchers to investigate other neurobiological processes as potential targets for therapeutic interventions in suicidal individuals. Since the 1993 report (4) of increased soluble interleukin‐2 receptor concentrations in plasma of suicide attempters, multiple studies have interrogated levels of pro- and anti-inflammatory cytokines in psychiatric disorders (5–10), non-fatal suicide attempts (11, 12), or in suicide decedents (13–15). In addition, studies have examined the relationship of psychiatric disorders to transformation of microglia from a relatively quiescent, or “resting” state, to the “activated” immune response state marked by changes in shape and function (16–21). Results are conflicting, with some finding increased pro-inflammatory cytokine secretion and microglial activation, and others observing the opposite, or failing to detect any change (8, 15, 22–26). Possibly, the discrepancies are in part the result of confounding factors intrinsic to human research (27, 28), but also, suicide probably involves biological features not found generally in psychiatric diseases or in nonfatal suicide attempts (29). Furthermore, autopsy studies differ intrinsically from studies in live subjects. Even after a severe suicide attempt, there is inevitably a delay before research studies can be performed; during that delay, the individual may emerge from a suicidal state. Postmortem studies of suicide capture brain biology at the moment of death from suicide.

In a previous study (25), we co-labeled paraffin sections for Iba-1 and CD68, a lysosomal marker, to distinguish activated microglia by their shapes and the abundance of CD68 immunoreactivity (IR). We compared prefrontal white matter from 11 individuals who died by suicide with that from 25 individuals with a similar distribution of psychiatric diagnoses who died in other manners. Iba-1-IR cells within or adjacent to vessel walls were broadly characterized as “perivascular cells,” with no classification regarding activation, since their morphological features were usually obscured by their close association with the walls of blood vessels. However, we found that densities of these perivascular phagocytes in DPFWM were greater in the suicide group than in the non-suicide psychiatric disorder group. A qualitative study of dorsal rostral cingulate white matter in suicide with major depression and non-suicide deaths without psychiatric disease came to a similar conclusion (30).

Typically, inflammation of the brain (e.g. in infection or demyelinating disease) involves changes in the blood-brain barrier (BBB) with an influx of peripheral immune cells (31–33), and there are reports of increased numbers of leukocytes in blood (34–36), cerebrospinal fluid (37, 38), and brains (39, 40) of individuals diagnosed with a psychiatric disease. Similarly, there is evidence of increased brain-immune interaction from animal studies of psychiatric disorders (41–43). We hypothesized that in suicide, excess Iba-1-IR phagocytes in or adjacent to the walls of blood vessels could represent increased trafficking of peripheral immune cells to the CNS or increased proliferation of resident perivascular macrophages (44).

Studies of human psychiatric disorders have focused more on microglia than on non-parenchymal monocytes and macrophages. In human postmortem work, distinguishing microglia from perivascular macrophages and peripheral monocytes has been problematic (45, 46). Markers typically used in animal research to distinguish CNS-resident from peripheral macrophages (CD45 or CD11b) are not appropriate for postmortem human work since they are also expressed by resident microglia (47) and infiltrating neutrophils and monocytes (48). We required a marker that would distinguish resident juxtavascular microglial cells from non-parenchymal macrophages and peripheral monocytes. While juxtavascular microglia can be touching a vessel wall or connected to it by a single process, with their cell bodies located outside the parenchymal basement membrane (49), non-parenchymal phagocytes are positioned between glia limitans and endothelial basement membranes (excellently illustrated in [50]). To distinguish between these 2 locations requires electron microscopy, but the cell types can be distinguished by staining for CD163 (ED2 in rats) (49), a macrophage scavenger receptor that binds haptoglobin-hemoglobin complexes following hemolysis (51). In human brain, CD163 is expressed by perivascular macrophages and by circulating or infiltrating monocytes or monocyte-derived macrophages (52–54). Microglial cells express CD163 only rarely, if at all (51).

To determine the contribution of perivascular or extracerebral macrophages to the increased perivascular cell density of Iba-1-IR cells in suicide, we quantified perivascular IR for CD163 in dorsal and ventral prefrontal white and gray matter. Since increased numerical densities of perivascular cells (number of cells per unit volume) could theoretically reflect increased vascularization in suicide, we also quantified surface area density of blood vessels (area of external surface of blood vessels per unit volume) in dorsal and ventral prefrontal white and gray matter.

MATERIALS AND METHODS

Human Brain Tissue

Procedures for obtaining and processing tissue and clinical data are as previously described in (25). Psychiatric diagnoses or their absence was determined by applying Diagnostic and Statistical Manual of Mental Disorders IV (DSM-ΙV) criteria (55), and the diagnosis was finalized at a consensus conference attended by senior clinicians and researchers of the Molecular Imaging and Neuropathology Division at NYSPI. Demographic data, clinical diagnoses, and mechanisms of death appear in Tables 1 and 2. Among suicide cases, one with schizoaffective disorder and one with bipolar disorder were included in the schizophrenia and mood disorder groups, respectively. Two non-suicides without schizophrenia or mood disorders had histories of alcohol dependence, one had an adjustment disorder, one had fully remitted bereavement, and one had a history of pathological gambling. Cases with active infections (<1% of collected cases) or grossly visible lesions in the frontal lobe (mostly gunshot wounds or acute contusions; <5% of collected cases) were excluded. However, occult neuropathological abnormalities are common from the sixth decade onwards, so to avoid bias, other histological lesions were not exclusionary.

TABLE 1.

Summary of Cases by Manner of Death

	Suicide (n = 11)	Non-Suicide (n = 25)	Test Statistic	p Value
Age (mean ± SD)	56 ± 18	55 ± 16	t(34) = −0.01	0.92
Sex	6 F/5 M	12 F/13 M	χ2(1, n = 36) = 0.1	0.72
PMI (mean ± SD)	12 ± 8	15 ± 7	t(34) = 1.1	0.28
pH (mean ± SD)	6.5 ± 0.2	6.3 ± 0.3	t(30) = −0.9	0.39
Psychiatric diagnosis	Schizophrenia (3)	Schizophrenia (7)	χ2(2, n = 36) = 5.5	0.05
	Affective disorders (5)	Affective disorders (3)
	No psychiatric illness (3)	No psychiatric illness (15)
Mechanism of death (n)	Respiratory (7)	Respiratory (4)	χ2(4, n = 36) = 10.5	0.03
	Traumatic (3)	Traumatic (11)
	Poisoning (1)	Poisoning (1)
		Cardiac (7)
		Other medical conditions* (2)

^*Other medical conditions included iatrogenic thrombocytopenia and acute hemorrhagic pancreatitis.

PMI, postmortem interval.

Table 1 reproduced from (25; Table 1).

TABLE 2.

Detailed Summaries of All Cases, Including Toxicology Results and Neuropathological Assessment

Case	Age	Sex	Suicide	Braak Stage*	Small Vessel Disease	Psychiatric Category†	Psychiatric Diagnosis	Blood Alcohol (g/100 mL)	Cotinine or Nicotine‡	Caffeine‡	Brain Toxicology	Iba-1 Vessel-Associated Cells per mm3§	Vascular Surface Density (microns2/mm3)§	Fraction of Vessel Wall Area Immunoreactive for CD163§	Cause of Death
1	35	M	No	Not determined	No	0	None	Not tested	Not tested	Not tested	Not tested	3358	9.12E+06	8.9%	Trauma (accident)
2	26	M	Yes	Not determined	No	0	Pathological gambling	0	Yes	No	Antidepressants	4120	9.93E+06	15.5%	Asphyxia by hanging (suicide)
3	42	F	No	Not determined	No	1	Schizophrenia. undifferentiated type	Not tested	No	No	Antipsychotics	2335	1.17E+07	29.5%	Thrombocytopenia
4	72	F	No	1	No	0	None	Not tested	No	Yes	Antihypertensives	4261	9.18E+06	20.9%	Heart failure
5	28	M	No	Not determined	No	0	None	0	Yes	Yes	Antidepressants	3850	9.93E+06	12.3%	Trauma (accident)
6	56	M	No	0	No	1	Schizophrenia	0	Yes	No	Benzodiazepines	4682	8.89E+06	12.0%	Myocardial infarction
7	56	M	No	1	No	1	Schizophrenia paranoid type	0	No	Yes	None	4117	1.06E+07	22.4%	Pulmonary tuberculosis
8	32	M	No	Not determined	No	0	None	0	No	No	None	5608	1.09E+07	19.1%	Gunshot wound (homicide)
9	58	M	No	6	No	0	None	0	Yes	No	Antidepressants	3534	1.22E+07	21.9%	Drowning (accident)
10	39	M	No	Not determined	No	0	Alcohol dependence in partial remission	0.22	No	Yes	None	3881	8.64E+06	13.8%	Trauma (accident)
11	66	M	No	1	No	0	None	0	Yes	No	Bronchodilators, benzodiazepines	2495	9.01E+06	5.6%	Trauma (accident)
12	66	M	No	0	Yes	2	Major depressive disorder, single episode, mild with melancholic features	0.048	No	Yes	Benzodiazepines	4106	9.90E+06	17.4%	Myocardial infarction
13	50	F	Yes	0	No	1	Schizoaffective disorder	0	Yes	Yes	None	3786	8.85E+06	10.5%	Trauma (suicide)
14	34	M	No	Not determined	No	0	Bereavement in full remission	0.046	Yes	No	None	3426	1.06E+07	25.1%	Trauma (accident)
15	72	M	No	2	Yes	0	None	0	No	Yes	None	4576	1.13E+07	6.3%	Exsanguination (homicide)
16	33	M	Yes	Not determined	No	0	Adjustment disorder with mixed anxiety and depressed mood	0	Yes	No	None	4337	8.37E+06	2.5%	Asphyxia by hanging (suicide)
17	51	F	No	0	No	0	None	0	Yes	No	None	4352	1.12E+07	9.8%	Trauma (accident)
18	71	F	Yes	2	No	2	Major depressive disorder, recurrent, most recent episode severe, with psychotic features	0	No	No	Antidepressants, benzodiazepines	4518	9.46E+06	17.0%	Drowning (suicide)
19	77	F	No	1	Yes	1	Schizophrenia undifferentiated	0.012	No	No	Hypnotics, antipsychotics	3130	1.21E+07	10.0%	Myocardial infarction
20	59	F	Yes	0	No	2	Major depressive disorder, recurrent, severe, without psychotic features,	0	No	No	Hypnotics, antidepressants, benzodiazepines	5888	9.79E+06	17.0%	Drowning (suicide)
21	42	F	Yes	0	No	2	Major depressive disorder. single episode, chronic, severe, with mood congruent psychotic features	0.006	No	No	Opioids, antidepressants, benzodiazepines	4944	9.32E+06	17.0%	Ingestion of acid (suicide)
22	68	F	No	1	Yes	2	Major depressive disorder. single episode chronic, melancholic, with moderate severity	0	Yes	No	None	3175	1.41E+07	12.3%	Myocardial infarction
23	29	M	No	Not determined	No	0	Alcohol dependence.	Not tested	Not tested	Not tested	Not tested	5585	9.35E+06	13.2%	Intoxication (accident)
24	79	M	Yes	1	No	2	Major depressive disorder, single episode	0	No	No	Benzodiazepines	5976	9.65E+06	17.8%	Asphyxia by hanging (suicide)
25	76	F	No	≤4	No	0	None	0	Yes	No	None	3317	1.03E+07	20.4%	Trauma (accident)
26	74	M	Yes	1	Yes	1	Schizophrenia paranoid type. dementia	Not tested	Not tested	Not tested	Not tested	3510	1.11E+07	17.0%	Asphyxia by hanging (suicide)
27	67	F	Yes	1	Yes	2	Bipolar disorder, most recent episode depressed, severe with psychotic features	Not tested	Not tested	Not tested	Not tested	4432	1.04E+07	20.4%	Trauma (suicide)
28	57	F	No	0	No	2	Major depressive episode	0	No	Yes	None	3740	1.07E+07	13.8%	Hemorrhagic pancreatitis
29	73	F	No	1	Yes	1	Schizophrenia undifferentiated	Not tested	No	No	None	5935	1.02E+07	19.1%	Bronchopneumonia
30	39	M	Yes	Not determined	No	0	None	0.282	Yes	Yes	None	4080	9.70E+06	28.8%	Asphyxia by hanging (suicide)
31	57	F	No	1	No	0	None	0	No	No	None	3985	1.19E+07	15.8%	Drowning (accident)
32	71	F	Yes	2	Yes	1	Schizophrenia paranoid type. dementia nos.	0	No	Yes	Antipsychotics, benzodiazepines, hypnotics	5928	8.63E+06	19.1%	Trauma (suicide)
33	55	F	No	0	No	1	Schizophrenia undifferentiated		Yes	No	Antipsychotics	4233	1.15E+07	16.2%	Myocardial infarction
34	67	F	No	0	Yes	0	None	0	Yes	Yes	Benzodiazepines	3301	1.01E+07	20.9%	Gunshot wound (homicide)
35	64	M	No	≤4	Yes	1	Schizophrenia paranoid type	0	Yes	No	Antidepressants, anticonvulsants	4712	1.14E+07	7.2%	Myocardial infarction
36	44	F	No	Not determined	No	0	None	0	No	Yes	NMDA antagonists, benzodiazepines	3488	1.02E+07	21.4%	Gunshot wound (homicide)

^*Not determined for individuals below age 45.

^†1, schizophrenia spectrum; 2, mood disorder; 0, neither.

^‡In brain.

^§In dorsal prefrontal white matter.

In brief, autopsies were performed within 24 hours of death at the Institute for Forensic Medicine in Skopje, Macedonia. Left hemispheres were sliced coronally at 2-cm intervals. The slices were fixed in 10% phosphate-buffered formalin for 5 days, and then rinsed and transferred to phosphate-buffered saline with 0.05% sodium azide and stored at 4°C. A coronal slice, approximately 4-mm thick was then taken at the level of the rostral tip of the frontal horn of the lateral ventricle (Fig. 1), divided into pieces of convenient size, embedded in paraffin, and cut at 6 microns onto positively charged slides (Fisher Scientific, Pittsburgh, PA). Clinical data were obtained in all cases by psychological autopsy interviews with relatives and, when available, review of psychiatric records.

FIGURE 1.

Coronal section of frontal lobe at the level analyzed. As indicated (horizontal line), we subdivided white and gray matter into dorsal and ventral regions at the dorsal edge of the corpus callosum (CC). Brodmann areas are noted for reference. Stock photograph showing representative anatomy; (25; Figure 1).

Immunohistochemistry

After deparaffinization and rehydration, sections were treated with 1% H₂O₂ (EMD Chemicals, Inc., Gibbstown, NJ) for 30 minutes. Antigen retrieval was performed by microwaving deparaffinized sections in 10 mM sodium citrate buffer, pH 6.0. The sections were then washed with phosphate buffer containing 0.1% Triton X-100 (Fisher Scientific, Hampton, NH), then incubated for 1 hour in 10% normal goat or horse serum (Fisher Scientific). The slides were incubated overnight at 4°C in mouse monoclonal anti-human CD163 IgG, clone EDHu-1 (56–58), 0.005 mg/mL. They were then washed, incubated with biotinylated anti-mouse IgG from horse (Vector Laboratories, Inc., Burlingame, CA), 7.5 μg/mL, washed, incubated with avidin-biotinylated peroxidase complex (Vector), washed, and incubated with 0.01% H2O2 (EMD Chemicals, Inc.), 0.02% 3,3′-diaminobenzidine-tetrahydrochloride ([DAB], Acros Organics, Morris Plains, NJ), yielding a brown reaction product. After washing, sections were incubated overnight at 4°C in a 1:2000 dilution of rabbit antiserum to Glut-1 (Abcam, Cambridge, UK) followed by alkaline phosphatase-conjugated anti-rabbit IgG and finally Permanent Red substrate (Dako, Carpenteria, CA), yielding a bright pink reaction product. Blood vessels (and erythrocytes) are stained pink and CD163 brown, allowing an unambiguous localization of CD163 in relation to the blood vessel wall (Fig. 2A, B).

FIGURE 2.

(A, B) Representative images of perivascular macrophages or circulating monocytes (see text) and blood vessels in human prefrontal white matter, immunostained for CD163 (brown) and Glut-1 (pink). CD163-IR monocytes are predominantly localized along the walls of blood vessels (black arrows) and have an elongated shape. However, some are rounder (black star). From case #29 (non-suicide). Scale bar = 50 μm.

Stereological Estimation of Blood Vessel Area Density and Area Density of Vessel-Associated CD163

The surface density of blood vessels was estimated stereologically. We used an Olympus BX 61 microscope (Olympus, Center Valley, PA) with a motorized 8-slide stage (Prior Scientific, Rockland, MA), a DP71 digital camera (Olympus, Center Valley, PA), with Visiopharm Software Version 3.6.5.0 (Visiopharm, Hørsholm, Denmark) image acquisition and NewCast stereology modules. The area of interest (white or gray matter) was outlined indirectly, using immediately adjacent paraffin sections stained for myelin with Verhoeff stain (59). To estimate vessel surface density, 438 × 330-μm fields of the region of interest (ROI) were defined at the intersections of a randomly placed grid with a line spacing of 1.5 mm in each direction. Each field was digitally photographed with a 20× objective (Olympus UPLFLN, N.A. 0.5), and a camera resolution of 2086 × 1572 pixels, yielding a nominal resolution of 0.21 microns per pixel.

A line probe was placed on the images to estimate the area densities of vessel walls and of CD163-IR blood vessel walls (60). Parallel test line segments, each associated with a point at one end, were uniformly superimposed on the digital images. The segments were 109.4 μm long, with 35 μm between adjacent parallel lines and 103 μm between the end of one segment and the beginning of the next collinear segment. The operator recorded the number of probe points falling on the ROI, and all intersections of the probe lines with the outer surface of a blood vessel within the ROI. Each intersection was defined as CD163-IR or CD163-negative. Surface area density (A_VTOTAL), which has the dimension of 1/length (area/volume), was calculated as: A_VTOTAL= 2 Σ (i)/[l Σ (p)], where i is the number of intersections of the outer surface of blood vessel with a line probe at each test site, l is the length of the line probe, and p is the number of points that fall onto the outlined tissue at each test site. CD163 area fraction (AF) was estimated as the ratio of CD163-IR intersections between vessel walls and line probes to the total number of intersections between vessel walls and line probes. Ratios were converted to their common logarithms before statistical analysis or plotting. Two observers (TS, IZQ), from whom demographic and clinical information were hidden, performed all counting of features in white matter and cortex, respectively.

Statistical Analyses

Statistical analyses were performed with IBM SPSS Statistics for Windows, version 24 (IBM Corp., Armonk, NY). As before, the values of the dependent variables were determined separately for dorsal and ventral white matter. Data were analyzed first by repeated measures ANOVA with 2 measures (blood vessel surface area density [A_VTOTAL] and CD163 area fraction [CD163 AF]), each with 2 levels (dorsal and ventral) in white and gray matter. The between-group factors were suicide (yes or no) and psychiatric group (schizophrenia, mood disorder, or non-psychiatric).

Secondary analyses employed one-way ANOVA, independent sample t-tests, or paired sample t-tests, as appropriate. Since data were not normally distributed for CD163 AF, we reanalyzed data using appropriate non-parametric tests (Kruskal-Wallis test and Wilcoxon Signed Rank test). All tests were 2-tailed.

RESULTS

Group Characteristics

Group characteristics were as previously described in (25). There were no differences in pH or PMI between groups defined by manner of death (suicide or non-suicide), psychiatric diagnosis, cause of death, or sex. Age was different between diagnoses, F(2, 33) = 4.7, p = 0.02, because the mood disorders group was older (63.6 ± 11.1) compared with those without psychiatric illness (47.7 ± 17.3). Males (49.2 ± 18.1) were younger than females (61.1 ± 11.8), t(34) = 2.3, p = 0.03. In addition, pH was inversely related to age (r = −0.42, p = 0.02). Toxicological findings are discussed below.

A_VTOTAL and CD163 AF by Manner of Death and Psychiatric Diagnosis

Stereological Considerations.

We postulated that perivascular CD163-IR cells would account for the difference between the densities of perivascular, Iba-1-IR cells in suicide and non-suicide decedents. This difference is approximately 700 cells per cubic mm of white matter, which is quite sparse: approximately 1.4 × 10⁶ cubic microns per cell (if regularly arranged, approximately 107 microns between cells). With 6-μm sections and a 20× objective, a convenient physical disector has a volume of 300 000 cubic microns, so a counting goal of 200 objects per ROI (the minimum needed for a coefficient of error of approximately $1/\sqrt{200 }$ = ∼0.07), would require evaluating approximately 1000 disectors per ROI for each case. This is impractical and inefficient, since, for this study, we are concerned only with the area comprising blood vessels and perivascular space. Furthermore, in these elongated cells, it is difficult to identify a consistent counting feature. Although the function of these cells, when apposed to blood vessels, is not known, it is reasonable to suppose that the extent of coverage of blood vessels is functionally at least as significant as the number of CD163-IR cells. We therefore used a line probe to determine the surface area density of the blood vessels in the white matter, and the fraction of the vascular surface area in contact with IR for CD163, assuming that an excess of CD163-IR cells adjacent to blood vessels would produce a larger fraction of vascular surface area in contact with CD163-IR components.

In white matter we counted (mean ± SD) 1428 ± 452 total intersections, 229 ± 108 intersections IR for CD163 (16%), and 5004 ± 1697 points, and 2217 ± 768 total intersections, 124 ± 72 intersections IR for CD163 (6%), and 3053 ± 1051 points in gray matter. Assuming a binomial distribution, the standard error for the AF of CD163 IR is 0.01 in the white matter, or 6% of the observed AF, and 0.005 in the gray matter, or 9% of the observed AF (see below). This was achieved with an average of 300 counting sites per ROI in the white matter and 254 counting sites in the gray matter, considerably more practical than the estimated 1000 sites per ROI that would be necessary to count 200 CD163-IR cells per ROI, assuming that the excess perivascular Iba-1-labeled cells are IR for CD163.

The theoretical standard error of the mean (SEM) of the stereological estimates of area surface density was computed (61) for each ROI (dorsal or ventral) in each case and divided by the mean for that ROI in that case to obtain a coefficient of error, which varied between 0.0015 and 0.01. The standard error of the mean for CD163 AF, without logarithmic conversion, was computed as:

$$\mathbf{S}\mathbf{E}\mathbf{M}=\sqrt[]{\frac{\mathbf{A}\mathbf{F}\left(1-\mathbf{A}\mathbf{F}\right)}{\mathbf{n}}}$$

and ranged from 0.005 to 0.018 in white matter (mean = 0.010), and from 0.004 to 0.005 in gray matter (mean = 0.003). SEM was divided by AF to obtain the coefficient of error, which ranged from 0.037 to 0.27 in white matter (mean = 0.067; 93% of values were below 0.10) and from 0.036 to 0.11 in gray matter (mean = 0.068). The measured coefficient of variation (standard deviation/mean; CV) = 0.36 for AF in white matter. From CV² = BV² + CE², where CV² is the measured variance, BV² is the true biological variance, and CE² is the estimation variance, we have BV = 0.35, 5.4 times greater than the estimation error in the white matter. In the gray matter, CV = 0.32, so BV = 0.31, 4.6 times greater than estimation error.

White Versus Gray Matter

Vascular surface area density (surface area of vessels in ROI/volume of ROI) was significantly higher in gray than in white matter (Fig. 3). Paired-samples t-tests showed that, A_VTOTAL was 59% higher in dorsal gray matter than in dorsal white matter, t(35) = −6.017, p < 0.001, and 63% higher in ventral gray than in ventral white matter, t(35) = −5.573, p < 0.001. CD163 AF was 64% higher in white than in gray matter, both dorsally, t(35) = 16.995, p < 0.001, and ventrally, t(35) = 12.897, p < 0.001 (Fig. 4A, B). These differences were independent of psychiatric diagnosis or manner of death (data not shown).

FIGURE 3.

Contrasting capillary density between human prefrontal gray matter and white matter (G and W, respectively, in minimized inset) immunostained for CD163 (brown) and Glut-1 (pink). From case #26 (suicide). Scale bar = 200 μm.

FIGURE 4.

Comparison of the total surface density of Glut-1-IR blood vessels (Av_TOTAL) (A) and the fraction of the vessel surface area immunoreactive for CD163 (CD163 AF) (B) in prefrontal white and gray matter. Paired samples t-tests show that dorsal Av_TOTAL was 59% higher in gray than in white matter (p < 0.001); ventrally the difference between vascularization in gray and white matter was 63% (p < 0.001). Av_TOTAL was significantly higher in white than in gray matter, irrespective of the manner of death or diagnosis (A). In contrast, CD163 AF was lower in gray than in white matter, both dorsally (p < 0.001) and ventrally (p < 0.001) (B). Symbols and bars represent mean ± SEM.

Effects of Suicide and Diagnosis

White and gray matter were analyzed separately. We performed repeated measures ANOVA with 2 measures (A_VTOTAL and CD163 AF), each with 2 levels (dorsal and ventral), 2-independent variables (suicide or not, and diagnostic group: schizophrenia, mood disorder, or none).

In the white matter, there was a significant effect of location, F(1, 30) = 6.92, p = 0.015, and a significant interaction of suicide and location, F(1, 30) = 4.77, p = 0.037, on A_VTOTAL. Consistent with these results, independent samples t-tests revealed that blood vessel surface area per unit volume of DPFWM was 10% smaller in suicide than in other manners of death [95% confidence interval = 3% to 16%; t(34) = 2.4, p = 0.018] (Fig. 5A). Furthermore, paired sample t-tests showed that in suicide, total vascular area density was 14% lower dorsally than ventrally, t(10) = 4.0, p = 0.002 (Fig. 5A), while in non-suicide deaths, the difference was only 4% and not statistically significant, t(24) = 1.1, p = 0.30. Non-parametric tests produced equivalent results. There were no significant effects of suicide on the fraction of vascular surface area IR for CD163 (Fig. 5B). Thus, the excess of vessel-associated phagocytes per unit volume in the suicide cases that we reported previously (25) cannot be attributed to an excess of vascular surface area per unit volume (there is actually a deficit), nor to an excess of vessel-associated peripheral macrophages. Comparisons between psychiatric groups revealed no significant effects of suicide, ROI, or interaction on A_VTOTAL (Fig. 6A) or CD163 AF (Fig. 6B) in the white matter.

FIGURE 5.

Association of suicide with the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (A) and the fraction of the vessel surface area immunoreactive for CD163 (CD163 AF) (B) in white matter. Suicide (purple) was associated with a lower Av_TOTAL in prefrontal dorsal white matter (p =0.018). Paired sample t-tests show that in suicide, Av_TOTAL was 14% lower dorsally than ventrally (p = 0.002), while in non-suicide deaths, the difference was only 4% and not statistically significant (p = 0.30) (A). No significant effects of suicide, region, or interaction were found in CD163 AF (B). Symbols and bars represent mean ± SEM.

FIGURE 6.

Association of psychiatric diagnosis with the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (A) and the fraction of the vessel surface area immunoreactive for CD163 (CD163 AF) (B) in white matter. Comparisons between diagnostic psychiatric groups revealed no significant effects of diagnosis, region, or interaction on Av_TOTAL (A) or CD163 AF (B). Symbols and bars represent mean ± SEM.

In the gray matter, we found a significant effect of location, F(1, 30)= 5.61, p = 0.025, on CD163 AF. Paired-samples t-tests revealed a trend toward significance for a higher CD163 AF in ventral gray than in dorsal gray matter, t(36) = −1.865, p = 0.071 (Fig. 7B). There were no statistically significant effects of manner of death (Fig. 7A, B) or psychiatric diagnosis (Fig. 8A, B) on any of the measures in gray matter.

FIGURE 7.

Association of suicide with the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (A) and the fraction of the vessel surface area immunoreactive for CD163 (CD163 AF) (B) in gray matter. No significant effects of suicide, region, or interaction were found on Av_TOTAL (A) or CD163 AF (B). Symbols and bars represent mean ± SEM.

FIGURE 8.

Association of psychiatric diagnosis with the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (A) and the fraction of the vessel surface area immunoreactive for CD163 (CD163 AF) (B) in gray matter. Comparisons between diagnostic psychiatric groups, revealed no significant effects of diagnosis, region, or interaction on Av_TOTAL (A) or CD163 AF (B). Symbols and bars represent mean ± SEM.

Correlations

Tables 3 and 4 summarize correlations between the new measures and those reported previously (25). In DPFWM, there was no significant correlation between density of perivascular Iba-1-IR cells and CD163-IR vascular AF, r² = 0.002. In the same area, vascularity and IR for CD163 were positively correlated with markers of microglial activation. A_VTOTAL and CD163 AF were positively correlated in dorsal gray matter and white matter (Table 5). Ventral white matter A_VTOTAL was significantly correlated with A_VTOTAL in gray matter (data not shown).

TABLE 3.

Correlation Coefficients Between Iba-1-IR Cells, Total Blood Vessel Surface Area Density and CD163 Vascular Area Fraction in DPFWM (Pearson’s r), n = 36

	Activated Phagocytes Dorsal	Resting Microglia Dorsal	Perivascular Cells Dorsal	Total Blood Vessel Surface Area density Dorsal
Total blood vessel surface area density dorsal	0.471**	−0.348*	−0.310	1
CD163 vascular area fraction dorsal	0.176	−0.402*	0.043	0.166

^**Correlation is significant at the 0.01 level (2-tailed).

^*Correlation is significant at the 0.05 level (2-tailed).

TABLE 4.

Correlation Coefficients Between Iba-1-IR Cells Total Blood Vessel Surface Area Density and CD163 Vascular Area Fraction in VPFWM (Pearson’s r), n = 36

	Activated Phagocytes Ventral	Resting Microglia Ventral	Perivascular Cells Ventral	Total Blood Vessel Surface Area Density Ventral
Total blood vessel surface area density ventral	−0.144	−0.095	−0.081	1
CD163 vascular area fraction ventral	0.205	−0.275	0.164	0.029

TABLE 5.

Correlation Coefficients for Total Blood Vessel Surface Area Density and CD163 Vascular Area Fraction in Prefrontal Dorsal White and Gray Matter (Pearson’s r), n = 36

	Total Blood Vessel Surface Area Density Dorsal White	Total Blood Vessel Surface Area Density Dorsal Gray	CD163 Vascular Area Fraction Dorsal White	CD163 Vascular Area Fraction Dorsal Gray
Total blood vessel surface area density dorsal white	1	0.438**	0.166	0.237
Total blood vessel surface area density dorsal gray		1	0.426**	0.480**
CD163 vascular area fraction dorsal white			1	0.622**
CD163 vascular area fraction dorsal gray				1

^**Correlation is significant at the 0.01 level (2-tailed).

Medications and Other Drugs

Toxicology for 179 drugs and metabolites, including all psychiatric medications and drugs of abuse commonly used in Macedonia, was performed by gas chromatography/mass spectroscopy on frozen cerebellar tissue, which was available from 32 of the cases (23 non-suicides, 9 suicides). Nine classes of medications were found in 13 combinations of 0–3 classes each (Table 2). The most commonly detected class of drugs was benzodiazepines (10 cases; 5 suicides), which are generally available without prescription in Macedonia. Antidepressant medications were found in 7 cases (4 suicides). Antipsychotic medications were found in 4 cases (1 suicide). A hypnotic (zolpidem) was found in 3 cases (2 suicides). An opioid (tramadol) was found in 2 cases (1 suicide). Anticonvulsant, antihypertensive, and bronchodilatory drugs, and the glutamate antagonist, phencyclidine, were found in one case each, all non-suicides. Alcohol levels were determined in postmortem blood from 31 cases; 26 were negative, and 5 (including 1 suicide) ranged from 0.046 to 0.28 g/dl.

Ignoring manner of death, in DPFWM we find benzodiazepines associated with lower values of A_VTOTAL, t(29.7) = 3.6, p < 0.001, whereas there is no significant effect of antidepressant or antipsychotic medications (data not shown). If we consider suicide and medication separately, values of A_VTOTAL in non-suicides without benzodiazepines were significantly greater than in suicides without benzodiazepines, t(20) = 2.8, p < 0.01, or non-suicides with benzodiazepines, t(21) = 2.5, p = 0.02 (Fig. 9A). Among suicides, there is no effect of benzodiazepines, and among those with benzodiazepines, there is no effect of suicide on A_VTOTAL. In other words, benzodiazepines and suicide are each associated with a statistically significant 15%–20% loss of DPFWM A_VTOTAL, but these associations are not additive (Fig. 9A). The lower value for DPFWM A_VTOTAL in suicide is not sensitive to antidepressant or antipsychotic drugs and is present in the absence of any drugs, t(12) = 2.4, p = 0.03 (Fig. 9B).

FIGURE 9.

Associations of the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (A, B) and density of vessel-associated Iba-1-IR cells (C, D) with suicide and benzodiazepines (A, C) or suicide and any drug detected on brain toxicology (B, D). p Values from independent 2-tailed t-tests. Symbols and bars represent mean ± SEM.

We performed a similar analysis of the association of medications with our earlier finding of greater numerical density of vessel-associated, Iba-1-IR cells in suicide (25). Suicide with benzodiazepines was associated with a significantly greater density than suicide without benzodiazepines, t(5.07) = 4.3, p = 0.008, or non-suicides with benzodiazepines, t(8) = 3.8, p = 0.005 (Fig. 9C). In other words, in addition to an effect of suicide, there is a significant interaction between benzodiazepines and suicide, which was confirmed by ANOVA (suicide: F(1, 32) = 8.3, p = 0.008; suicide × benzodiazepines: F(1, 32) = 7.1, p = 0.01; benzodiazepines: not significant). Densities of vessel-associated phagocytes were not significantly associated with antidepressant or antipsychotic medications. The medicated suicide subjects had significantly greater vessel-associated phagocyte density than did medicated non-suicides, t(16) = 3.9, p < 0.001, or unmedicated suicides, t(6.68) = 3.2, p = 0.02. In the group without drugs, there was no effect of suicide (Fig. 9D).

There were no significant associations of caffeine with any outcome measure of this study. There were no significant associations between drugs and CD163 AF in any region. Detection of nicotine or its more stable metabolite, cotinine, was associated with a 15% lower average value of perivascular phagocyte numerical density in DPFWM, t(25.9) = 2.2, p = 0.04 (data not shown).

Neuropathology

We perform routinely a neuropathological examination on the left cerebral hemisphere of each case. Slides were stained with hematoxylin and eosin and, and if the deceased was age 45 or older, with AT8 antibody to hyperphosphorylated tau protein. AT8 staining is extremely sensitive for the detection of neuritic senile plaques, neuropil threads, and neurofibrillary tangles. Braak and Braak (62) stage was determined for each such case. In 2 cases, there was too little hippocampal tissue to determine the extent of involvement, but the absence of neocortical involvement allowed us to make a rating of ≤4. The ratings are included in Table 2. There were no significant associations of Braak score with any of the outcome measures. Only one brain, from an individual without history of dementia, contained pronounced Alzheimer-type pathology. Eliminating this case from the analysis did not change the classification of any of the outcome measures as significant or not.

Small vessel disease (SVD) was commonly present in the older individuals. Significant small vessel disease was limited to those who died at age 60 or older; of these, it was present in 60% of the suicide cases and 70% of the non-suicide cases (Table 2). Varying combinations of tortuosity, thickening, and hyalinization of small vessels were present. We found no significant effect of SVD on any if the outcome measures. Although not statistically significant, among the brains from individuals aged 60 or over, vascular area densities in dorsal white matter, ventral white matter, dorsal cerebral cortex, and ventral cerebral cortex were 7%–22% greater in individuals with SVD, which is consistent with the qualitative neuropathological features of the vessels.

Inspection of Results

Even though the difference between suicide and non-suicide deaths is statistically significant, there is overlap between the 2 groups. In our studies, we found suicide to be characterized by low vascular area density and high density of perivascular Iba-1-IR cells in DPFWM. Since these 2 measures are only weakly correlated, r = −0.31, p = 0.07, we examined these variables in 2 dimensions to look for atypical cases, dividing the space into quadrants delineated by the median values of the 2 variables (Fig. 10). The upper left quadrant (high Iba-1vascular cell density, low vascular area density) contains 64% of the suicide cases and 12% of the non-suicides, while the lower right quadrant contains 9% of the suicide cases and 40% of the non-suicide cases. However, we found no clinical, demographic, neuropathological, or toxicological features to differentiate the suicide case in the lower right quadrant or the 3 non-suicide cases in the upper left quadrant from the other suicide and non-suicide cases, respectively.

FIGURE 10.

Plot of density of vessel-associated Iba-1-IR cells (vertical axis) against the total surface area density of Glut-1-IR blood vessels (Av_TOTAL) (horizontal axis), in dorsal prefrontal white matter. Heavy black lines represent medians for the entire sample. Numbers labeling symbols correspond to case numbers in Table 2. Symbols and bars represent mean ± SEM.

DISCUSSION

Recently, we reported the association of suicide with greater densities of Iba-1-IR cells adjacent to blood vessels in DPFWM, compared with non-suicide deaths of individuals with a similar distribution of sex, age, and psychiatric conditions (25). We considered several possible explanations of this finding. Increased density of perivascular immune cells could reflect increased trafficking of peripheral monocytes toward the BBB in suicide. Evidence of stress-induced influx of peripheral immune cells into brain comes from animal models of social defeat stress, where maladaptive behavior is associated with perivascular accumulation of peripheral immune cells and their infiltration into the brain beyond the perivascular space (63, 64). In that model, the effect of stress could be mitigated by deactivating IL-1R on endothelial cells (65), suggesting that stress can attract peripheral inflammatory cells to the brain by inducing endothelial cells to express an inflammatory profile of cytokines.

Alternatively, the number of non-parenchymal perivascular macrophages could increase by their proliferation or migration along the abluminal surface of blood vessels, much as the cells from the yolk sac presumably populated the perivascular spaces during early development (44). A third possibility is the migration of macrophages from the brain parenchyma (microglia), meninges, or choroid plexus (meningeal or choroid plexus macrophages) (44) either into or adjacent to the perivascular space. Finally, the excess of vessel-associated macrophages could represent an excess of blood vessels (either more or larger) in suicide.

To estimate the contribution of non-parenchymal immune cells to increased densities of Iba-1-IR vessel-associated cells and to measure vascular surface area density, we conducted stereological estimates of vascular surface area density (A_VTOTAL) by staining for glucose transporter Glut-1, and the fraction of vascular surface area IR for CD163 (CD163 AF) in dorsal and ventral prefrontal white and gray matter. This study employed the same paraffin blocks from the same cases that we used in the earlier study (25).

We found no association of suicide with extent of perivascular IR for CD163. Furthermore, there was no significant correlation between CD163-IR, in or around vessel walls, and density of vessel-associated Iba-1-IR phagocytes. All cases manifested CD163-IR. The IR cells had spheroidal bodies and long processes that were often apposed to the abluminal surface of small blood vessels. CD163-IR somas or processes covered approximately 16% of the surface area of parenchymal blood vessels. The antibody labeled no other structures. We found no classical balloon-shaped cells in any of the study cases, although IR clusters of such cells were common in an unrelated case of vasculitis, processed in parallel. In summary, there was no evidence of quantitatively or qualitatively abnormal IR for CD163 in or around blood vessels, and no correlation between perivascular CD163-IR and perivascular Iba1-IR. Since we would expect CD163-IR in perivascular macrophages, meningeal macrophages, and choroid plexus macrophages (53), the excess perivascular cells IR for Iba-1 and CD68 in suicide are most likely microglia. Finally, the possibility that excess of vessel-associated macrophages represents an excess of blood vessels (either more or larger), can be discounted, since we found that the vascular surface area density in DPFWM was actually lower in suicide.

We found that A_VTOTAL in the cerebral cortex is approximately 60% greater than in white matter in both dorsal and ventral prefrontal regions. Greater vascularity of gray matter than white is well known from histological (66–68) and MRI perfusion studies (69). However, the CD163 AF is approximately 3 times greater in white than in gray matter, both dorsally and ventrally.

We determined the quantity A_VTOTAL to test whether there was an increase in blood vessel surface area per unit volume that accounted for the increased number of perivascular phagocytes per unit volume in DPFWM in suicide. As noted above, this possibility was excluded. However, in DPFWM, we found statistically significant associations of suicide with lower A_VTOTAL. The functional significance of lower A_VTOTAL in this ROI is unclear. It could reflect regional cerebral blood volume, which would suggest a local decrease in metabolic activity. While not achieving statistical significance in any of our 4 ROI’s, in each ROI we found greater A_VTOTAL in brains with SVD than in brains of similarly aged people without SVD. Among all individuals without SVD, the association between suicide and dorsal white matter A_VTOTAL was statistically significant, p = 0.007. In brains with SVD, the effect of suicide was similar in magnitude but not statistically significant, probably reflecting the small number of brains with SVD, as well as greater heterogeneity in vascular morphology in those brains. Reduced perfusion of the prefrontal cortex (70, 71) and reduced blood levels of vascular endothelial growth factor (VEGF) (72, 73) are reported in people who attempted suicide. We doubt that such associations are related to our observation of reduced A_VTOTAL in suicide. We did not find differences in A_VTOTAL in the cortex, nor in VPFWM, whereas imaging studies show predominantly cortical differences. Similarly, one would expect the effects of circulating VEGF to be more diffuse than the changes we observe. Perhaps it should not be surprising that our results differ from those of studies of live individuals. Even if those subjects eventually died by suicide, they were not studied at the time of death. We studied brains that, aside from postmortem changes, were in the same state that they were in during the fatal event.

We also found effects of medications detected by postmortem toxicology on the brain. The results are complicated, because, while only 3 categories of drugs (benzodiazepines, antidepressants, and antipsychotics) were commonly found, they were present in various combinations of these classes, and in even more combinations of individual medications. Nonetheless, we found a striking association of benzodiazepines with lower A_VTOTAL and greater density of vessel-associated phagocytes in DPFWM. Although suicide is associated with lower DPFWM A_VTOTAL in the absence of drugs, there is a similar association between lower DPFWM Av_TOTAL and benzodiazepines in the absence of suicide. Furthermore, the increase in Iba-1-IR cells in suicide is not found in the group without benzodiazepines. These effects could be mediated in part by the mitochondrial translocator protein, also known as the peripheral benzodiazepine receptor. These receptors serve many functions, including the regulation of microglial and astrocytic reactivity (74), and they are expressed in brain endothelial cells (75). Their possible involvement in the regulation of vascularity is supported by a similar effect on A_VTOTAL in VPFWM, p = 0.03, also seen only in the non-suicide group. The potency and direction of these effects are largely unexplored and very probably differ among various benzodiazepines, as well as among various types and functional states of phagocytes. Another possibility is that both the presence of benzodiazepines and the act of suicide are indicators of current stress, and that stress alters vascularization (76).

Thus, we showed in this study that the increased density of vessel-associated immune cells in suicide cannot be attributed to the proliferation of non-parenchymal perivascular macrophages nor to the influx of peripheral immune cells toward the BBB, unless these cells lack the usual IR for CD163 (44). We also showed an association of suicide with smaller vascular surface area density in DPFWM but not in VPFWM or prefrontal cortex. Increased presence of immune cells at the BBB and changes in vascularization point to possible alterations in the properties of the neurovascular unit in suicide, which requires further study.

Limitations

This study, combined with our previous study of the same material, relied on immunohistochemistry for just a few antigens: Iba-1, CD68, Glut-1 and CD163. It would, of course, be desirable to confirm and refine the results with additional markers, as is done routinely in studies of rodents. However, this study used human autopsy tissue, and we chose the most robust and well-established staining methods that we could find. As noted above, we employed CD163 area fraction instead of CD163 cell density because the time to count a sufficient number of perivascular CD163-IR cells would have been prohibitive; likewise, we could not devote the time required for stereological studies with equivocal markers. It is possible that there are Iba-1-IR perivascular cells that are not IR for CD163 but are not microglia. However, except for CD163, we are not aware of definitive positive markers to distinguish microglia from non-parenchymal perivascular macrophages or monocytes in the human brain. TMEM119 (66) and the purine receptor, P2Y12 (68), have been used to study human brain, but their specificity is not certain.

The small sample places some limits on statistical power. We found effects of suicide with no effect of psychiatric diagnosis, but for a given effect size, we had greater statistical power to detect an effect of suicide than an effect of diagnosis. Our intent was to study suicide, not specific psychiatric diagnoses. In any event, we had sufficient power to accept the null hypothesis with confidence and to reject our hypothesis that the additional perivascular phagocytes in suicide are IR for CD163, since we were examining sections from the same paraffin blocks in which we found the excess perivascular phagocytes, and furthermore, we found no correlation between numerical density of perivascular Iba-1-IR cells and the area fraction of CD163-IR vessel walls. The small size of the sample does become a problem when attempting to subdivide the groups by medications or SVD, for instance. One alternative is to screen all cases beforehand and to use only those that are free of medications and neuropathology, but since these are common, such screening is likely to result in an idiosyncratic sample. Automation of stereological counting is not yet routine or accepted, but progress in that direction is being made, with the aim of reducing counting time ten-fold (77). That will greatly increase the feasibility of using redundant markers and analyzing more samples.

We examined only the left cerebral hemisphere because the right cerebral hemisphere was sliced and rapidly frozen and could not give comparable histology. A final limitation is that we did not adjust the results for tissue shrinkage during fixation and embedding, and we did not use isotropic sectioning. The CD163-IR fraction of vessel surface area would not be affected by these omissions, because they would presumably have an equal effect on vessel surface area with or without IR for CD163. While the measurements of vascular surface area density may well be biased (i.e. the values would probably be different if obtained with anisotropic sampling [see (78)] or correction for shrinkage), the comparison of dorsal and ventral prefrontal white matter in the same brain slice should not be affected. We found such a difference (dorsal A_VTOTAL< ventral A_VTOTAL) only in the suicide group. Within-subject differences that are unique to one group should help to design additional studies with the enhanced statistical power that comes from having each brain serve as its own control.

This study was supported by grants DIG-0-041-13 and PDF-129-15 from the American Foundation for Suicide Prevention, grants MH064168, MH098786, and MH090964 from the National Institutes of Health, and gifts of software and equipment from Visiopharm and Olympus.

The authors have no duality or conflicts of interest to declare.