The Search for Pain Biomarkers in the Human Brain

André Mouraux; Gian Domenico Iannetti

Disclosures

Brain. 2018;141(12):3290-3307. 

In This Article

Abstract and Introduction

Abstract

Non-invasive functional brain imaging is used more than ever to investigate pain in health and disease, with the prospect of finding new means to alleviate pain and improve patient wellbeing. The observation that several brain areas are activated by transient painful stimuli, and that the magnitude of this activity is often graded with pain intensity, has prompted researchers to extract features of brain activity that could serve as biomarkers to measure pain objectively. However, most of the brain responses observed when pain is present can also be observed when pain is absent. For example, similar brain responses can be elicited by salient but non-painful auditory, tactile and visual stimuli, and such responses can even be recorded in patients with congenital analgesia. Thus, as argued in this review, there is still disagreement on the degree to which current measures of brain activity exactly relate to pain. Furthermore, whether more recent analysis techniques can be used to identify distributed patterns of brain activity specific for pain can be only warranted using carefully designed control conditions. On a more general level, the clinical utility of current pain biomarkers derived from human functional neuroimaging appears to be overstated, and evidence for their efficacy in real-life clinical conditions is scarce. Rather than searching for biomarkers of pain perception, several researchers are developing biomarkers to achieve mechanism-based stratification of pain conditions, predict response to medication and offer personalized treatments. Initial results with promising clinical perspectives need to be further tested for replicability and generalizability.

Introduction

Physical pain is intrinsically unpleasant and aversive. This is the very reason why it is advantageous for survival: it drives behaviours that avoid bodily injury when interacting with the environment. Yet, especially in modern societies, acute pain is often devoid of behavioural advantage. Think, for example, of the pain experienced during medical interventions. Furthermore, an increasing number of individuals suffer from pain that lasts for months or years (Breivik et al., 2006). This chronic pain is not only lacking any obvious behavioural benefit (Hodges and Tucker, 2011), but also heavily impairs quality of life. The fact that pain has a major negative impact on human wellbeing is often used as a persuasive argument to justify the funding of pain research. The prospect, which has shaped the way many pain neuroscientists conceive, design and interpret their work, is that beyond improving basic knowledge of the neural mechanisms of sensory perception, research in the field of pain will also lead to the development of more effective means to treat pain and reduce suffering.

Three arguments are usually set forth to uphold this prospect in the field of pain neuroimaging (Kupers and Kehlet, 2006; Borsook et al., 2007, 2010; de Vries et al., 2013; Lee and Tracey, 2013; Morton et al., 2016; Grosen et al., 2017; Tracey, 2017). First, it is often claimed that functional neuroimaging could be used to derive brain biomarkers that measure pain 'objectively'. This would provide a solution to the hurdle of assessing pain using verbal reports, which are considered to be inherently prone to response biases (Wager et al., 2013; Kumbhare et al., 2017). Such biomarkers for pain would make it possible to quantify pain severity and the effects of treatments in an objective and undisputable 'evidence-based' fashion. Second, it is postulated that a mechanism-based diagnosis of clinical pain conditions is essential for adequate pain management (Woolf and Max, 2001; Woolf, 2008; Borsook et al., 2010, 2011; Lee and Tracey, 2013). By disclosing the neural mechanisms underlying pain in individual patients, neuroimaging could thus improve clinical diagnosis and care, for example, by predicting individual response to treatment (Wartolowska and Tracey, 2009; Denk et al., 2014; Tetreault et al., 2016; Davis and Seminowicz, 2017; Kumbhare et al., 2017). Third, it has been proposed that functional neuroimaging and electrophysiology could be used to quickly identify new pain-relieving drugs by characterizing their effects on CNS pain 'circuits' (Woolf and Max, 2001; Martucci et al., 2014), an approach sometimes referred to as 'pharmaco-fMRI' or 'pharmaco-EEG' (Schweinhardt et al., 2006; Wise and Tracey, 2006; Woolf, 2008; Gram et al., 2013).

One important question challenges the use of functional neuroimaging to derive 'biomarkers' of pain perception: does the brain activity sampled by these techniques when an individual experiences pain correspond to the neuronal activity causing the emergence of the painful percept? As summarized in a review paper that we published a few years ago (Iannetti and Mouraux, 2010), we and others (Carmon et al., 1976; Chapman et al., 1981; Melzack, 1999; Downar et al., 2003) have expressed concern regarding the specificity for pain of the brain responses classically observed when experiencing transient pain, i.e. the so-called 'pain matrix', a label covertly implying some specificity for pain. The concern is based on the observation that largely the same functional neuroimaging responses can be elicited by non-painful stimuli, provided that they are salient enough (Chapman et al., 1981; Downar et al., 2003; Mouraux and Iannetti, 2009; Mouraux et al., 2011) (Figure 1). More recently it was also shown that a virtually identical 'pain matrix' response can be observed in patients with congenital insensitivity to pain (Salomons et al., 2016), thus providing further evidence that these brain responses are largely non-specific for pain. (This statement does not imply that neural activities specific for pain do not exist. Instead, it implies that the neural activities captured by current EEG or functional MRI techniques, which reflect synchronous activity within large populations of neurons, are—at the very least—largely unspecific for pain.) To escape from these controversies, many researchers now refrain from using the term 'pain matrix', and opt instead for terms like 'pain network', 'pain signature' or 'neural circuits' (Tracey and Mantyh, 2007; Seifert and Maihofner, 2011; Lelic et al., 2012; Longo et al., 2012; De Simone et al., 2013; Wager et al., 2013). Such labels are equally suggestive of the idea that the brain responses that are being measured reflect neural activity somehow unique for pain. To elaborate on only one of these examples, the term 'signature' denotes a distinctive pattern, product or characteristic by which something can be unequivocally identified. As detailed below, we argue that the attempts to falsify the hypothesis that the brain responses being measured are specific for pain using appropriate control stimuli have been insufficient, and the liberal use of terms implying specificity has biased the interpretation of several pain neuroimaging results.

Figure 1.

Transient nociceptive stimuli causing pain. In this example heat laser pulses delivered to the right hand (scalp EEG and intracerebral LFP) or foot (functional MRI) elicit large-scale brain responses. In scalp EEG, the response is dominated by a large negative-positive wave maximal at the scalp vertex (electrode Cz), probably originating from bilateral operculo-insular regions, the cingulate cortex and, possibly, the contralateral primary somatosensory cortex. Responses in similar regions are also detected using functional MRI. Importantly, equally salient but non-painful and non-nociceptive tactile or auditory stimuli elicit very similar EEG and functional MRI responses, indicating that most of this activity is unspecific for pain or nociception and, instead, multimodal (Mouraux and Iannetti, 2009; Mouraux et al., 2011). Similarly, although the insula has been proposed to be strongly involved in pain perception, equally salient nociceptive and non-nociceptive stimuli trigger similar local field potentials (LFPs) recorded directly within the insula (Liberati et al., 2016). Nevertheless, other less prominent features of the sampled activity might be more selective for pain or nociception, as reflected by the selective increase of gamma-band oscillations (GBOs) when painful heat stimuli are presented (Liberati et al., 2018). BOLD = blood-oxygen level-dependent; ER% = event-related change in oscillation amplitude; ERP = event-related potential.

The danger of assuming that brain responses sampled when experiencing pain are specific for pain is well illustrated by the way several pain neuroimaging results have been communicated by the general media. For example, a press release reporting a neuroimaging study on pain in infants conducted by Goksan et al. (2015) stated that because the 'brains of babies light up in a very similar way to adults when exposed to a painful stimulus, new-borns experience pain in the same way as adults' (http://www.ox.ac.uk/news/2015-04-21-babies-feel-pain-'-adults'). Evidently, this conclusion, based on reverse inference, is valid if and only if the observed brain activity is specific for pain, as detailed in the 'Pain-specific and pain-selective brain activity' section below.

In the following sections, we first examine whether an established assumption—that there is a real clinical need for an 'objective' laboratory measure for the subjective perception of pain—is truly justified. Second, we examine the issue of pain specificity of the brain activity sampled using functional neuroimaging and electrophysiological techniques. This is necessary and timely, given the increasing use of new methods to analyse brain activity such as multivariate pattern analysis of functional MRI data to reveal 'pain signatures' (Wager et al., 2013), as well as the proposal of new theoretical concepts such as the 'pain connectome' (Kucyi and Davis, 2015, 2017), in which pain would emerge from widespread brain network activity. Third, we assess pragmatically whether current biomarkers derived from neuroimaging have the ability to measure pain 'objectively'. Finally, we evaluate the strength of the evidence supporting the use of functional neuroimaging to perform mechanism-based stratification of patients with chronic pain, predict response to treatment, and assist the pharmacological development of novel treatments for pain.

processing....