46 Neuroimaging of suicidal behaviour: Where does the field stand?

Consistent evidence implicates serotonin system dysfunction in the neurobiology of suicidal behaviour. Neuroimaging studies link brain structure and function in vivo and contribute to our understanding of neural pathways. Areas of the prefrontal cortex and limbic structures are targeted in neuroimaging studies of suicidal behaviour, which have focused on structural, haemodynamic, metabolic, and neuroreceptor changes in the brains of suicide attempters. Neuroimaging studies have revealed that signal hyperintensities, perfusion and metabolic abnormalities, processing of affect and serotonin receptor and transporter changes, may each play a role. Knowledge regarding the neurobiology of suicidal behaviour must rely on study designs utilizing robust methodologies, including improved patient and control group selection, improved neuroimaging techniques, and adequate statistical analysis to enhance the validity, consistency, and conclusiveness of the data. Ongoing development of new radioligands and imaging methodologies promise to enhance our ability to delineate the neurobiology of suicidal acts.

There is consistent evidence implicating serotonin system dysfunction in the neurobiology of suicidal behaviour. In early studies by Åsberg and colleagues (1976), depressed suicide attempters were found to have lower cerebrospinal fluid (CSF) serotonin metabolite, 5-hydroxyindoleacetic acid (5-HIAA), than depressed non-attempters. High-lethality suicide attempters had lower CSF 5-HIAA levels when compared to those who used a low-lethality method (Mann and Malone 1997; Placidi et al. 2001). Through post-mortem studies of suicide victims, the dorsal raphe nucleus and the ventral prefrontal cortex (PFC) have been identified to have serotonin abnormalities such as decreased serotonin transporter (SERT) binding and increased serotonin (5HT)-1A receptor binding (for a review see Mann et al. 1999).

Neuroimaging studies provide the opportunity to link brain structure and function in vivo and can contribute to our understanding of neural pathways in suicidal behaviour. Based on post-mortem findings and current understanding of serotonin pathways in the brain, areas of the PFC and structures of the limbic system are targeted in many neuroimaging studies of suicide (Oquendo and Mann 2001). Using available imaging modalities, these studies have focused on structural, haemodynamic, metabolic, and neuroreceptor changes in the brains of suicide attempters.

In order to identify relevant neuroimaging studies of suicidal behaviour for this review, we conducted a search in PubMed using the following key words: suicide (or suicide attempt or suicidal behaviour) and MRI (magnetic resonance imaging)—or fMRI (functional magnetic resonance imaging) or diffusion tensor imaging (DTI) or PET (positron emission spectography) or single photon emission computerized tomography (SPECT). All articles identified were reviewed, and those with original data regarding suicidal behaviour were included. Those articles that did not specifically address the topic of suicidal behaviour were excluded from this review. In addition, bibliographies were scanned to identify additional relevant publications not found with our search strategy.

We identified thirteen studies using this strategy, with significant variations in study design and methodology. Data from each study is presented in Table 46.1 to highlight design, number of study participants, imaging modality used, major findings reported by the authors, and major limitations.

The thirteen neuroimaging studies highlighted in Table 46.1 suggest that underlying brain abnormalities reflected by signal hyperintensities, perfusion or metabolic abnormalities, functional differences in processing of affect, and serotonin receptor and transporter changes may play a role in suicidal behaviour. Although these results are intriguing, due to differences in study design, most of them cannot be directly compared. It is also difficult to draw firm conclusions.

Several methodological recommendations may improve the likelihood of determining the neurobiological basis of suicidal acts. These include improved sample selection, imaging technique, and statistical analysis. Below, we detail strategies that may improve our fund of knowledge regarding the biological basis for suicide attempts.

From a sampling point of view, future studies would benefit from utilizing a standardized definition of suicidal behaviour, well-matched controls, as well as sufficient subjects to ensure the statistical power to detect clinically relevant abnormalities.

To date, imaging studies have used a variety of definitions regarding suicidal behaviour, as researchers differ on the definition of a suicide attempt. One widely accepted definition of a suicide attempt is the ‘potentially self-injurious behaviour with non-fatal outcome for which there is evidence that that the person intended at some level, to kill himself’ (O'Carroll et al. 1996). This definition involves three components: self-injury, non-fatal outcome, and intent to die as a consequence of the behaviour. However, some studies have included patients with a history of deliberate self-harm, irrespective of the intent of the act. In other studies, descriptions are not detailed enough to permit a clear definition of suicide attempt for comparison to other studies. Such inclusion criteria discrepancies can create potential confounders when interpreting results. The definition of suicide attempt is important, not only because we seek to elucidate the underlying biological basis of as uniform a phenotype as possible, but also because there is evidence that the degree of suicidal intent and lethality of suicide attempt correlate with neurobiological markers such as tryptophan uptake and cerebral perfusion in the PFC (Oquendo et al. 2003; Leyton et al. 2006).

The use of appropriate controls groups is another key sampling issue. Although many of the studies have used healthy volunteers as controls, this type of matching does not permit full characterization of the abnormalities that are specific to suicidal behaviour. The inclusion of psychiatric control groups may help clarify the role of underlying psychopathology in observed abnormalities. Thus, well-matched, well characterized, control groups that take into account drug abuse history and comorbid psychiatric conditions, may clarify the confounding effect of these factors in the development of abnormalities observed in neuroimaging studies. Similarly, the method of suicide attempt used by the subject may have key effects on observed neuroimaging abnormalities.

Patient inclusion criteria based on time from suicide attempt varies in the studies reviewed, and points to an underlying question of whether propensity for suicidal behaviour and associated biological findings are enduring traits. Indeed, some studies image patients within an average of 1 or 2 weeks post suicide attempt (Audenaert et al. 2001, 2002; van Heeringen et al. 2003; Leyton et al. 2006), while others imaged patients with a lifetime history of suicide attempt, even if the attempt was over 4 years in the past or time from suicide attempt was not reported (Ahearn et al. 2001; Oquendo et al. 2003; Monkul et al. 2007; Jollant et al. 2008). Suicidal behaviour may be a state in which brain changes involved are only detectable at the time of the suicide attempt and given the plasticity of the brain, there are advantages to studying suicide attempters close to the time of the attempt. However, serious methodological issues must be grappled with. For example, acute and long-term effects of drug overdose on brain chemistry, use of alcohol during the attempt, and relative hypoxia experienced during certain types of suicide attempts can affect the central nervous system. Of course, damage from suicide attempts can be long lasting and, thus, even those studies of attempters with a lifetime history of suicidal behaviour may be affected by the consequences of the suicidal acts themselves. For these reasons, these issues require consideration during the design phase of the study.

There are several factors that may improve the validity of the data acquired in neuroimaging studies and neuroreceptor studies in particular (Audenaert et al. 2001; Parsey et al. 2002; Oquendo et al. 2003; Lindstrom et al. 2004; Leyton et al. 2006; Cannon et al. 2006; Oquendo et al. 2007). First, initial studies should be conducted with full quantification of binding potential through the use of metabolite corrected input functions. The use of arterial input functions or arterialized venous input functions should be standard practice until reference regions are determined to be measures of free and non-specific binding and are found to be invariant between suicide attempters and non-attempters. Once it is determined in a sample of sufficient size that the reference region is adequate, future studies can be carried out with reference tissue modelling approaches.

Second, choice of PET as the imaging modality would lead to improved data collection, since PET imaging offers much higher resolution than current commercially available SPECT imaging systems. Given the small volume of some structures of interest, the higher resolution of PET offers a distinct advantage.

Third, each year we see the development of new and improved radioligands for use in human studies. State of the art radioligands ideally have good blood–brain barrier permeability, stable tracer kinetics, absence of radioactively labelled metabolites that have affinity for the target, and measurable free fraction. Key considerations also include ease of production and a favourable safety profile in humans. For example, [¹¹C]DASB is an excellent alternative to the previous serotonin transporter radioligand [¹¹C]McNeil 5652. Namely, it has a measurable free fraction (Ogden et al. 2007), kinetics that are more amenable to accurate modelling, and it generates a higher signal to noise ratio. Utilization of superior radioligands will improve the accuracy and reliability of the data collected.

Our literature search revealed few blood flow studies and only one metabolism study and one functional imaging study (Audenaert et al. 2002; Oquendo et al. 2003; Fountoulakis et al. 2004; Jollant et al. 2008). In order to assess functional differences and establish the presence or absence of a functional neurocircuitry of suicidal behaviour, these studies are essential. Again, initial studies should be conducted with metabolite input functions as there may be global changes in flow or metabolism that are not detectable using qualitative scans.

To analyse neuroimaging data, statistical parametric mapping (SPM) software is commonly used, but it requires care in interpretation. This is due to the issue of multiple comparisons. While the program has the capability to perform analyses to test the difference in regions of interest (ROI) that are defined a priori, most commonly, the statistical approach used is not hypothesis-driven. In this type of exploratory analysis, the statistical program makes inferences about differences in signal across the brain. Specifically, it morphs individual brains to fit a standardized brain space and creates a three dimensional matrix or ‘map’, also referred to as a ‘glass brain’. The program assigns a value to individual volumes in the space referred to as voxels. By literally calculating millions of t-tests, voxel by voxel across the entire brain volume, it makes estimates about group differences. It generates two different types of output. First, it identifies points in the brain where the signal from the imaging methodology is significantly different from other regions. These brain locations are identified by the p value calculated for that specific point in the ‘glass brain’ but the p value, referred to as an ‘uncorrected p value’ in the output, is not corrected for multiple comparisons. Then, utilizing sophisticated statistical theory, SPM generates two additional types of output. One type of output is an estimate of the probability that a given cluster of adjacent or colocalized voxels, that together exceed a threshold in terms of a predetermined size of the cluster, is a region of significant difference between groups. This cluster is assigned a p value. SPM also generates ‘corrected p values’ for points within the significantly different cluster that are significantly different in the groups being compared. The threshold for significance can be set by the user. Although this powerful, data-mining methodology can be useful for hypothesis generation, it requires that the output be interpreted within its limitations. For example, not every p value that is ‘significant’ is meaningful (uncorrected p values) because the problem of multiple testing confounds the results. However, as mentioned, the program does generate p values that are ‘corrected’. These are the p values that are most reliable. Uncorrected p values or analyses conducted in which the threshold for generating images that demonstrate regions of ‘significant difference’ is too low, and cannot be interpreted meaningfully.

In general, there are several key challenges in the interpretation of neuroimaging studies of suicidal behaviour. For example, some structural studies have focused on presence of signal hyperintensities (Ahearn et al. 2001; Erlich et al. 2005). However, the presence or location of signal hyperintensities is not specific to suicidality. Signal hyperintensities are regions of decreased regional cerebral blood flow (Wen et al. 2004) reflecting areas of vascular, neuronal, or other brain parenchymal damage. They are associated with ageing, geriatric depression, dementia, cardiovascular disease, methamphetamine, cocaine and opiate abuse (Greenwald et al. 1998; Barber et al. 1999; Breeze et al. 2003; Lyoo et al. 2004; Bae et al. 2006). Nonetheless, the studies of signal hyperintensities are instructive since key brain pathways involved in mood regulation and behavioural inhibition may be affected. Signal hyperintensities may disrupt neural communication between PFC, limbic and other pivotal regions. They may result in impaired impulse regulation and other functions central to decision-making, which may impact on the emergence of suicidal acts.

Similarly, interpretation of studies using C-α-methyl-L-tryptophan is hampered by studies that question the use of labelled tryptophan in CNS as a marker of serotonin synthesis (Leyton et al. 2006). Indeed, in anesthetized rhesus monkeys, labelled tryptophan appeared to be a marker of tryptophan uptake, rather than serotonin synthesis (Shoaf et al. 2000). In light of such disagreement, caution must be used in interpretation of results using this radioligand.

Among the newer modalities for neuroimaging, diffusion tensor imaging (DTI), may help elucidate neural substrates for suicidal behaviour. While it is a new technology in schizophrenia research (Hoptman et al. 2002; Kanaan et al. 2005), DTI may provide a measure of white matter integrity and, therefore, an assessment of the degree of connectivity among brain regions involved in mood regulation and executive functioning. DTI uses MRI scanners and has two common outcome measures, fractional anisotropy and probabilistic tractography, or trace. Fractional anisotropy reflects the direction of diffusion of water between nerve tissue, with lower values reflecting axonal disorganization. Trace is the diffusion coefficient across all directions, with higher values reflecting increased extracellular space, suggesting an abnormality (Hoptman et al. 2002). Potentially, this technology could more specifically identify regions of neural disconnection which may contribute to manifestation of suicidal behaviour.

There is much to learn from in vivo neuroimaging studies of suicidal behaviour. Neuroimaging studies have revealed that factors such as signal hyperintensities, perfusion and metabolic abnormalities, and serotonin receptor and transporter changes, may each play a role in suicidal behaviour. As discussed, the advancement of knowledge regarding the neurobiology of suicidal behaviour relies on the design of studies utilizing robust methodologies, including improved patient and control group selection, improved neuroimaging techniques, and adequate statistical analysis to further enhance the validity, consistency, and conclusiveness of the data. The ongoing development of new radioligands and imaging methodologies are likely to enhance our ability to uncover the underlying neurobiology of suicidal acts.