The mapping of cortical activation by near-infrared spectroscopy might be a biomarker related to the severity of fibromyalgia symptoms

The delta value of oxyhemoglobin (Δ-HbO) determined by functional near-infrared spectroscopy at prefrontal cortex (PFC) and motor cortex (MC) based on primary (25 °C) and secondary (5 °C) thermal stimuli presented a larger peak latency at left MC in fibromyalgia than in controls. The difference between HbO concentration 15 s after the thermal stimuli ending and HbO concentration before the thermal stimuli onset (Δ-HbO*) at left PFC increased 47.82% in fibromyalgia and 76.66% in controls. This value had satisfactory discriminatory properties to differentiate cortical activation in fibromyalgia versus controls. A receiver operator characteristics (ROC) analysis showed the Δ-HbO* cutoffs of − 0.175 at left PFC and − 0.205 at right PFC offer sensitivity and specificity of at least 80% in screening fibromyalgia from controls. In fibromyalgia, a ROC analysis showed that these cutoffs could discriminate those with higher disability due to pain and more severe central sensitization symptoms (CSS). The ROC with the best discriminatory profile was the CSS score with the Δ-HbO* at left PFC (area under the curve = 0.82, 95% confidence interval = 0.61–100). These results indicate that cortical activation based on Δ-HbO* at left PFC might be a sensitive marker to identify fibromyalgia subjects with more severe clinical symptoms.

Experimental paradigm. Subjects sat in a comfortable chair with their arm rested, lights turned off, and room temperature kept constant (around 21 °C). After explaining, demonstrating, and addressing concerns about each thermal stimulation, we asked subjects to keep their eyes closed and to reduce any kind of motor activity not related to the experiment. After verifying the correct positioning and adequate signal capture, we proceeded to data registering. Preliminarily we recorded 60 s in a resting state, during which the participant was requested to stay relaxed with open eyes, fixing on a point on the computer monitor. Thermal stimulation was based on cold pressor test and its capacity of reproducing tonic pain. It was performed while recording the cortical activation for posterior offline analysis. For the thermal test, the participants immersed their right hand in a bucket with water at two different temperatures: 25 °C (innocuous-primary stimulus) and 5 °C (noxious-secondary stimulus). They maintained their hand immersed for 30 s or until feeling the first pain sensation. After each thermal test (primary or secondary), the participant rested for 2 min. A digital thermometer measured the water temperature during all experiment time. Thus, we reduce the possible confound effect of different temperatures and other environmental stimuli that could produce cortical activation and influence in our experiment. The total time of the experimental paradigm was around 15-20 min. Figure 2 illustrates the paradigm used.
Assessment of pain disability due to pain and central sensitization symptoms. 0-100) assessing physical symptoms, emotional distress, headache/jaw symptoms, and urological symptoms. This scale allows a rapid tracking of symptoms associated with central sensitization to guide therapeutic strategies and indicate prognostic factors. Higher scores indicate a higher degree of self-reported symptomatology. Additionally, part B of SCI assesses the presence of psychiatric diagnosis and neurological disorders associated with central sensitization 30 .

Assessment of sociodemographic characteristics and other clinical and psychological measures. The American
College of Rheumatology 2 criteria was used in physician-administered and patient self-administered questionnaires, increasing the correct diagnoses. The scale of FM symptoms ranging from 0 to 3 and adding the widespread pain index (WPI) to the modified severity scale (SS scale). The WP quantifies the extent of bodily pain on a 0-19 scale by asking patients if they have had pain or tenderness in 19 different body regions (shoulder girdle, hip, jaw, upper arm, upper leg, lower arm, and lower leg on each side of the body. As well as the upper back, lower back, chest, neck, and abdomen) over the past week, with each painful or tender region scoring one point. The CSS scale quantifies symptom severity on a 0-12 scale by scoring problems with fatigue, cognitive dysfunction, and unrefreshed sleep over the past week, each on a scale from 0 to 3. Also, it assesses if there are problems at the last 6 months often present, such as life-disturbing problems, pain or cramps in the lower abdomen, depression symptoms, headache, etc. We used the Beck Depression Inventory (BDI-II) 31 for depressive symptoms, the State-Trait Anxiety Inventory-short version (STAI-SV) 3 , and the Brazilian Pain Catastrophizing Scale (BP-PCS) 32  Data analysis. All continuous variables were tested for normality using the Shapiro-Wilk test and box-plot diagrams. The majority presented an indication of non-parametric distributions. Therefore, for sample characterization, we used Mann-Whitney U tests for continuous variables and Fisher's exact tests for categorical variables. We compared the effect of thermal stimuli through the speed on reaching the HbO peak (seconds) and by the difference of Δ-HbO pre-to post-stimuli within in each group (fibromyalgia or controls) and between groups 33 . Generalized Estimating Equations (GEE) models with an exchangeable working correlation were used to compare the group effect (fibromyalgia and controls) speed of activation (peak latency) and to compare the group effect on cortical deactivation based on Δ-HbO* value. In the models, interactions among the factors (group and temperature) were also examined. Effect sizes were reported according to Cramer's V [for one degree of freedom (df), the effect size is classified as small (0.1), medium (0.3), or large (0.5)] 34 . Spearman's rho coefficient was used to assess the relationship between left and right PFC activation assessed by the Δ-HbO* with the disability due to pain and central sensitization symptoms in fibromyalgia. After confirming the corresponding assumptions, a multivariate linear regression model was performed to adjust for multiple to assess the relationship between dependent variables [scores related to disability due to pain for daily activities (BP-PCP:S) and central sensitization scores (CSI-BP)] and the cortical activation level assessed by Δ-HbO* as the independent variable. All analyses were adjusted for multiple comparisons using Bonferroni's multiple comparison test. The area under the curves (AUCs) with exact binomial 95% confidence intervals (CI) are presented. The cutoff values with the highest Youden index, with 80% sensitivity and 80% specificity, are presented in each of four indexes, and all showed a receiver operator characteristics (ROC) AUC higher than 0.68.
For the analysis of an association between the cortical activation assessed by Δ-HbO* on PFC and fibromyalgia symptoms, a priori sample size estimation indicated a study of 20 subjects for type I and II error rates of 0.05 and 0.20, respectively, and anticipating an effect size of 0.6 for multiple regression analysis, which allows for two predictors (cortical activation level assessed by the Δ-HbO* on the left and the right PFC). Finally, considering the likely attrition rate and other unexpected factors, we increased the sample by 10%, and the required sample size was 22 patients 35 . All analyses considered a significance level of α < 0.05 for two-tailed tests. To analyze the data, we used the software SPSS version 22.0 (SPSS, Chicago, IL, USA).

Results
Sample characterization. A total of 46 volunteers were enrolled in the study, and 5 were excluded (1 from the fibromyalgia group due to problems with the fNIRS registration and 4 from the control group due to acute illness and use of antidepressants). The comparisons between groups related to sociodemographic and clinical characteristics are presented in Table 1. Groups differed significantly in all characteristics, in the sense that fibromyalgia subjects were older, with higher body mass index and fewer years of study. As expected, the fibromyalgia group showed more severe clinical symptoms and higher medication use. Data are presented in Table 1. The GEE models revealed significant main effects for the group in the maximum cortical activation (peak latency) of HbO curve at left MC (Wald χ 2 = 5.39, df = 1, P = 0.02, Cramer's V = 0.36) ( Table 2). However, we found no difference in either temperature effects or interaction between temperature and group. For the fibromyalgia group, the peak latency difference at primary thermal stimulus (25 °C) versus secondary thermal stimulus (5 °C) was an increase equal to 15.50% [mean (standard deviation [SD]) 5.61 (0.52)/6.48 (0.62)] compared to 1.11% in controls [mean (SD) 8.09 (0.90)/8.00 (0.73)], respectively. Therefore, cortical activation occurred at a slower speed in fibromyalgia patients than in controls.
The GEE models revealed a significant temperature effect in Δ-HbO* at both PFC and MC. The main effect indicated there is an increment on Δ-HbO* at 5 °C comparing to 25 °C on both PFC and MC. For the fibromyalgia group at the left PFC the Δ-HbO* difference at primary thermal stimulus (25 °C) versus secondary thermal stimulus (5 °C) was an increase equal to 47.82% [mean (SD) 0.23 (0.03)/0.34 (0.05)] compared to 76.66% in fNRIS measures as a marker of cortical activation indexed by Δ-HbO*. Cortical deactivation based on Δ-HbO* at either the left or right PFC with their respective cutoff points reached at least 85% sensitivity and 80% specificity in the AUC analysis to discriminate fibromyalgia from controls. Data are shown in Table 3.

Analysis of the relationship between fibromyalgia symptoms and cortical activation after thermal stimuli.
Cortical deactivation based on Δ-HbO* values at PFC and MC and their correlations with fibromyalgia symptoms are presented in Table 4. Scatter plots of correlations between the Δ-HbO* values at PFC with scores in the CSS and disability due to pain are presented in Figure 4 of supplementary material.
Multivariate analysis of the relationship between the severity of fibromyalgia symptoms and PFC assessed by Δ-HbO*. Cortical deactivation based on Δ-HbO* value showed a statistically significant correlation with scores of disabilities due to pain and central sensitization (Table 4). Therefore, we examined their relationship by Generalized Linear Models (GLM). Dependent variables were disability due to pain and central sensitization scores, and the independent variable was Δ-HbO* as a measure of cortical activation at  www.nature.com/scientificreports/ both sides of the PFC. The results of these adjusted multivariate models are presented in Table 5. At the left PFC, a lower Δ-HbO* concentration was positively correlated with disability due to pain and central sensitization symptoms. In contrast, at the right PFC, a lower Δ-HbO* concentration was conversely associated with disability due to pain and central sensitization symptoms.  www.nature.com/scientificreports/ Δ-HbO* based on HbO before thermal stimuli onset and 15 s after thermal stimuli onset at PFC distinguishes patients with more disability due to pain and more severe CSS. We conducted a ROC analysis, stratifying for the cutoff points on Δ-HbO* to differentiate fibromyalgia subjects from controls, − 0.175 at the left PFC and − 0.20 at the right PFC (Table 3). The sensitivity, specificity, and AUC using these cutoff points to screen subjects with higher central sensitization symptoms and higher disability due to pain are presented in Table 6.

Discussion
This study's main findings highlight that the temperature effect produced a more considerable difference in the absolute concentration of HbO measured by Δ-HbO* at the left PFC in controls compared to fibromyalgia. In contrast, this difference at the left MC was more significant in fibromyalgia than in controls. We found this difference in Δ-HbO* pattern at the left PFC had satisfactory discriminatory properties to differentiate cortical activation in fibromyalgia patients versus controls and discriminate fibromyalgia subjects with more severe CSS symptoms and disability to pain. Also, the peak latency difference within-group revealed that the cortical activation occurred slower at the left MC in fibromyalgia than in controls. In sum, results from the present study suggest that the dynamic measure of HbO changes indexed on peak latency of HbO and Δ-HbO* are sensitive inferential markers to identify cortical dysfunction related to fibromyalgia. These findings contrast with our initial hypothesis that peak latency and differences in HbO concentration before and after thermal stimuli would be shorter and larger, respectively, in fibromyalgia subjects than in controls exposing that fibromyalgia group would have faster and stronger cortical response. The initial hypothesis was based on the rationale that sustained chronic pain could increase the excitability of pain pathways. The corresponding functional finding would indicate hyperactivation in target areas either as involved in pain processing or used as therapeutic targets to improve pain measures, nominally PFC and MC. Although we found that both cortical activation measures indicated hypoactivation in the left PFC and the left MC in fibromyalgia, the results from the current study are consistent and add information to suggesting that inferential measures based on BOLD signal might be valuable tools to differentiate dysfunctional cortical processing, mainly in areas associated with cognitive and emotional aspects of fibromyalgia, nominally left PFC. Table 4. Spearman correlation coefficients of the relationship between fibromyalgia symptoms with the delta value of oxyhemoglobin (ΔHbO*) as cortical deactivation measure in either PFC or MC (n = 22). Prefrontal cortex (PFC); motor cortex (MC). Correlation is significant at the 0.01 level (2-tailed); ΔHbO* in millimolar (mM). **Correlation is significant at the 0.05 level (2-tailed www.nature.com/scientificreports/ The variability of the BOLD signal that we found in our present study is indirectly coupled with neurovascular changes attributed to the intrinsic variability of neural processing, such as synaptic transmission or neurotransmitter functions indirectly assessed by HbO concentration 36 . Such results are not to replace existing diagnostic tools but rather to establish a meaningful neurobiological basis for cortical pain processing to open a new avenue to study the relationship between dysfunction in the neural networks in these areas. In this case, PFC and MC have been used as therapeutic targets for the application of transcranial stimulation, such as tDCS and repetitive TMS. Thus, this model is potentially valuable for differentiating patients from controls or predicting severity of symptoms. However, it is unclear whether it is a model of the relevant clinical pathology at all. Hence, it needs further prospective testing on independent data, since our primary goal is not to provide a complete model of symptoms and behavior but to test hypotheses about structure-function associations based on a welldefined experimental paradigm. Even though the current findings need further validation in longitudinal studies and larger samples, they extend the evidence that fibromyalgia features deteriorated function according to HbO changes across contrasting thermal stimuli. Following this perspective, the larger peak latency difference found in the left MC in fibromyalgia subjects compared with controls might be related to the reduced tone of cortical motor areas. Even though our experimental condition does not assess motor performance, our results agree with the literature showing reduced information processing speed in MC areas in fibromyalgia patients 37,38 . Thus, this result might indicate dysfunctional cortical processes related to fear of movement found in patients with chronic pain 39 or cognitive problems with impaired motor processing 37,38 . An alternative explanation could be based on a partial inhibition of cortical motor areas during concurrent nociceptive stimulation, which generates a loss of the metabolic advantage in consonance with the severity of symptoms. Our finding related to maximal activation and average changes in HbO in fibromyalgia subjects compared to controls agrees with a previous study that found lower maximal and mean changes in HbO concentration at both the left and the right PFC in fibromyalgia patients compared to healthy controls during a breath-holding task 40 . Also, another study using fNIRS found reduced brain activity over the frontal regions during a verbal fluency test (VFT) in fibromyalgia patients compared to controls 41 . In earlier studies, a more significant short intracortical inhibition was found in fibromyalgia patients than healthy subjects 4 . Thus, this set of findings suggests reduced PFC activation in fibromyalgia may indicate deterioration in cortical processing function. Although the size of effect related to the group and temperature on cortical activation is only moderate 42 , they support the experimental paradigm used to activate one or more brain target regions involved in pain processing, nominally PFC and MC. These results based on cortical differences of HbO offer a model with consistent predictive properties to discriminate fibromyalgia subjects from controls and to identify fibromyalgia subjects with more severe symptoms, however, caution is warranted before generalizing these findings, since the sample size is small and only one experimental paradigm was tested.
Aligned with this perspective to comprehend the physiology of cortical dynamic processing, the relevance of these findings is to show a framework for understanding the relationship between the function of cortical areas in pain processing and clinical symptoms related to CSS and disability due to pain. In this regard, they open an avenue to identify measures with the potential to be biomarkers, since they integrate the function of cortical areas involved in underlying fibromyalgia symptoms. According to experimental studies, these cortical areas (i.e., PFC and MC ) have been targets to improve pain by non-invasive brain stimulation techniques such as repetitive TMS 43 and tDCS [44][45][46][47] . This effect was showed with extended home-based use of tDCS on the left dorsolateral PFC (DLPFC), which improved pain, psychological symptoms, sleep quality, and disability due to fibromyalgia 44 . Also, previous studies demonstrated that the use of tDCS on the left DLPFC improved attention, working memory, and pain in fibromyalgia 48 and depressive symptoms in major depressive disorders 49,50 . In the same way, another study with healthy males using an electrical standardized stimulus on the unilateral right accessory spinal nerve unilateral showed in an integrative framework measures that the fNIRS measures Table 6. ROC analysis to screening the severity of fibromyalgia symptoms according to the right and left PFC activation based on Δ-HbO* (n = 22). AUC area under the curve, CI confidence interval, ROC receiver operator characteristics, mM millimolar. www.nature.com/scientificreports/ are suitable to comprehend the connections of PFC with the MC in pain processing 51 . Despite limitations in comparing results in these studies because of differences in sex and type of stimulus used in the experimental paradigm, these findings corroborate the role of functional spectroscopic mapping of pain-processing cortical areas related to sensory-discriminative and affective-motivational pain dimensions. Likewise, results of a metaanalysis of studies employing experimental pain stimuli indicate a positive association of the following brain areas with pain processing: primary and secondary somatosensory cortices, insular cortex, ACC, PFC, and thalamus 48 .
Thus, these results find support from the anatomic perspective, since the discrimination of pain intensity by the ventral pathway activates the PFC bilaterally, and the spatial discrimination by the dorsal direct path from the posterior parietal cortex triggers the DLPFC activation 40 .
A relevant contribution of our findings is to add results to literature integrating neurophysiological measures with clinical data with the perspective of accelerating the translation of surrogate measures to results to apply at the bedside. The differences in the concentration of Δ-HbO* at the left PFC was positively correlated with disability due to pain and CSS. In contrast, the change in Δ-HbO* at the right PFC by the same thermal stimulus was conversely correlated with the severity of these clinical symptoms (see Table 5). This suggests an imbalance of inter-hemispheric activation, and based on this experimental paradigm, the hypoactivation tends to be more pronounced in the left, whether at MC or PFC. One hypothesis to explain these findings is that some target areas are activated to the detriment of other circuits' deactivation. This imbalance may be related to functional lateralization of the amygdala in the context of pain. In general, the amygdala's right central nucleus tends to have a pronociceptive role, while the left has an antinociceptive role 52 . One hypothesis is that the disturbed balance of this system results in chronic pain conditions. Although the explanation for this imbalance is not clear, it may be due to the maintenance of neuronal hyperexcitability in the central left nucleus of the amygdala to counteract the pain-driving effect of the right central nucleus of the amygdala 52 . However, studies related to amygdala lateralization in the context of pain are only beginning, and there are many different elements to consider that will vary significantly based on pain mechanism. We acknowledge that this hypothesis remains relatively broad, but a better comprehension of lateralization of pain processing is relevant to the therapeutic perspective mentioned above. However, the specific intricacies necessary to fully understand how lateralization functions in pain will need further studies that address the side as a variable.
The ROC analysis was used to screen the severity of fibromyalgia symptoms observed after activation of the left PFC based on Δ-HbO*. The cutoff was defined by setting the AUC to offer 100% sensitivity and 98% specificity to screen fibromyalgia patients versus controls (Table 2). It offered an AUC of 0.82 to screen for the severity of CSS. As mentioned above, this result can be a consequence of lateralization of the antinociceptive response. This more considerable change of cortical response in fibromyalgia suggests a deteriorated function of the PFC in pain processing. More precisely, from a conceptual perspective, our findings might explain the pathophysiological processes that underlie fibromyalgia since they integrate the severity of symptoms with dysfunctional changes in cortical areas involved in the cardinal sign of PFC dysfunction, such as cognitive impairment 53 . Additionally, PFC dysfunction might explain the central sensitization-related clinical variables, such as depressive symptoms, insomnia symptoms, perceived level of disability 54 , duration of pain, current pain intensity 55 , average pain intensity, and pain catastrophizing 56 .
We assumed cortical activation changes could not be interpreted as a direct response to nociceptive stimuli since the hemodynamic response is an alternative marker of neuronal activity. Even though we cannot interpret them as a cause-effect response, their contextualized interpretation can help elucidate cortical brain function in targeting areas involved in pain processing and regions targeted by therapeutic approaches. Our findings agree with some of the literature supporting fNIRS measures as a reliably sensitive central measure to comprehend the cortical effect of painful stimulation 36 . Yücel et al. (2015) found that in healthy subjects, the BOLD signals assessed by fNIRS detected a greater activation in M1 upon painful stimulus but no critical changes in PFC. Although many studies have investigated cortical activation using fNIRS in fibromyalgia, a novelty of the current study is the distinct paradigm used 22,57 . Although these results indicate changes by indirect means (metabolic and vascular), they give us new insights to study in real time the effect of different paradigms to evaluate cortical dysfunction related to chronic pain. Additionally, they may be an entry port to assess the complex pain-related neural network and to understand the role of PFC and MC in processing pain signals.
Although these results offer a perspective to comprehend the role of target areas related to pain and emotion using fNIRS that permits an assessment in real time, we need parsimony in their interpretation owing to some limitations in the methods and study design. First, it is a challenge to maintain fibromyalgia patients in the same position for any length of time, and they may even experience scalp pain during hair manipulation. For these reasons, we removed some channels in some cases, and sometimes the quality of the signal was poor. However, we know these difficulties are intrinsic to this type of measure. Second, we identified differences between groups related to age, years of formal education, psychiatric disorders, and medication, several of which are expected. Even though some confounding effects cannot be fully controlled, we found results that indicate differences between groups with biological plausibility. Third, this is a physiological-basis study involving cortical pain processing and given the known difference between sexes in pain processing, we included only female subjects. We understand that this restricts external validity. However, it permitted us to reduce the potential confounding effect of sex on our measures. This is plausible since women are more susceptible to negative emotional responses such as fear of pain 58 , stress, and anxiety 59 . Fourth, we did not find an interaction between group and temperature. Although the explanation is not clear, it might be explained as an error type II. And it is not accessible to compared results among studies that used distinct paradigms to evoke the cortical activation because the kind of stimulus and its intensity is vital to the recruitment of cortical areas. It is important to realize that the cold pressure test is an acute stressor used to measure pain threshold (i.e., first feeling pain). Finally, further research is needed to assess cortical brain activation in chronic pain under different experimental paradigms, such as www.nature.com/scientificreports/ behavioral tests to determine a cognitive and emotional response, motor tasks, and changes related to effects of different therapeutic approaches (e.g., tDCS, TMS, etc.).
In conclusion, these results indicate that cortical deactivation based on Δ-HbO* at PFC might be a sensitive marker to discriminate fibromyalgia cortical processing and to screen patients with more disability due to pain and more severe CSS. Overall, they offer insight into cortical function in the pathophysiology of primary chronic pain and the modifications of the cortical part of these target areas in response to an effective treatment.