A cross-sectional analysis of brain structure, pain behaviors, and mental health in persons with surgically confirmed endometriosis

In this cross-sectional study, participants were females assigned at birth and were aged 12–44 years old. All participants were premenopausal. Two individuals identified as non-binary or gender fluid. There were seven pediatric participants under the age of 18. Forty-three participants (including six pediatric participants) with surgically confirmed endometriosis (SCE) comprised the exposed group, and twenty-six pain-free participants (including one pediatric participant), who were never diagnosed with endometriosis (NDE) and who were not undergoing clinical assessment for endometriosis, comprised the unexposed group. There is a possibility that some of those never diagnosed with endometriosis without a history of pelvic pain could have undiagnosed and/or asymptomatic endometriosis; however, this would be only one or two study participants, and their misclassification as unexposed to endometriosis would drive the results toward the null 30 . Everyone in the endometriosis group had surgical confirmation of the disease, which requires laparoscopy with or without histological studies of the tissue 31 , 32 . Participants with SCE were mainly recruited from academic medical centers around the Longwood Medical Area by study physicians (RA and ML). Both SCE and NDE exposure groups were recruited through advertisements at Boston Children’s Hospital (BCH) internal web server and the local Boston community (posters and word of mouth). Inclusion criteria included being right-handed to minimize bias in brain imaging data (see Jang et al. 33 ; Tomasi and Volkow 34 ) and having a surgically confirmed diagnosis of endometriosis or being scheduled for the surgery (and then diagnosed). Exclusionary criteria included recreational drug use, pregnancy, MRI-incompatible implants, claustrophobia, comorbid chronic pain conditions with pain intensity greater than that attributed to endometriosis, history of chronic pain (pain for more than 3 months), or significant medical history besides endometriosis. Exclusionary medical history included self-reported asthma, traumatic brain injury, any significant neurological injuries or disorders, history of psychosis or schizophrenia, severe personality disorders, and bipolar disorder. Participants were excluded if they currently took opioid analgesics. All potential participants were screened over email or over the phone for eligibility. If eligible, participants attended one study visit at our research and imaging center. The same exclusionary criteria were applied to the group with no diagnosis of endometriosis; however, inclusionary criteria also included no history of acute or chronic pain or any medical condition with significant CNS influence. To minimize the influence of hormonal milieu variability, all participants had no history of hysterectomy or oophorectomy, were not in menopause (defined as cessation of menses for 12 consecutive cycles unrelated to exogenous hormonal influences) and were between days 2 and 10 of their menstrual cycle on the day of MRI. This is consistent with previous literature in analyzing the impact of endometriosis on neuroimaging markers 27 . Participants taking hormonal contraceptives completed the study at any point in their cycle. There was no significant difference in current use of hormonal contraceptives across groups ( x 2  = 1.27, p  = 0.26). All procedures and protocols were approved by the BCH Institutional Review Board, as well as a reliance agreement with Brigham and Women’s Hospital, and participants and legal guardians provided informed consent and/or assent (for those under 18 years of age) before participating in the study. All ethical regulations relevant to human research participants were followed. After informed consent and assent (for those under 18 years of age) procedures, participants completed a battery of self-report questionnaires. General health history and menstrual screening forms were completed using RedCap 35 , 36 to provide demographic (i.e., race, ethnicity, gender, education level) and clinical information (i.e., menstrual cycle information, presence or absence of pain, neurological, psychiatric disorders, endometriosis diagnosis details). The Endometriosis Impact Questionnaire 37 (EIQ) was administered to participants in the SCE group to measure the long-term impact of endometriosis on different aspects of their lives over the last 12 months relative to their study visit. The questionnaire comprises 63 questions that measure the impact of endometriosis on physical, psychological, social, sexual, fertility, and employment factors. This questionnaire was demonstrated to have very good internal consistency reliability for the total questionnaire (Cronbach’s alpha = 0.98) and for each recall period of the EIQ (Cronbach’s alpha = 0.97 for last 12 months, 1–5 years ago, more than 5 years ago) 38 . The EIQ was also found to have a statistically significant intra-class correlation between all dimensions at times 1 and 2 (ICC. = 0.88–0.99) and good concurrent validity with a statistically significant positive correlation between the last 12 months of the EIQ and the Endometriosis Health Profile-5. A second study found the EIQ to have good measure quality and utility  39 . A brief battery was administered to each participant that included self-report questionnaires that measure cognitive appraisal and evaluation of pain. Participants reported their current pain score or average pain intensity on the day of the visit, and their pain score during menstruation using a 10-point Likert scale (0 = no pain at all to 10 = worst pain imaginable). In this study, we designed the eligibility criteria to exclude controls with any history of chronic pain states and recruited individuals with endometriosis that had higher, although varying, pain experiences. Intrinsically, this allows for a low pain state represented by the healthy patient group, and a higher pain state represented by the endometriosis group. The endometriosis group had a baseline higher pain level, high enough to elect for surgical intervention. The Pain Sensitivity Questionnaire (PSQ), a 17-item survey, measures pain perception and sensitivity based on painful daily situations 40 . The PSQ asks participants to imagine various painful daily life situations and rate their pain on a scale of 0–10, with 0 being no pain and 10 being the worst pain imaginable. The total score for each participant was calculated as the average rating of questions 1–8; 10–12; 14–17 40 . The score excludes questions which are not normally rated as painful by healthy subjects (i.e., “shaking hands with someone”). Scores can range from 0 to 10, with greater scores reflecting greater general pain sensitivity. The PSQ had acceptable reliability in our sample (Cronbach’s α  = 0.774), Pain catastrophizing or pain-related worry was captured by the Pain Catastrophizing Scale (PCS) 41 , with adolescents completing the adapted Pain Catastrophizing Scale-child (PCS-C) 42 . Participants respond to 13 statements describing thoughts and feelings associated with pain. Total scores range from 0 to 52 for both adult and pediatric questionnaires, reflecting the degree of catastrophizing and worry. For both adult and pediatric surveys, scores below 15 indicate low catastrophizing, scores between 16 and 23 represent moderate catastrophizing, and scores higher than 24 suggest high catastrophizing 41 , 43 . In our sample, the PCS had questionable reliability (Cronbach’s α  = 0.658). The Pain Anxiety Symptom Scale (PASS), a 20-item measure, was also utilized to capture pain-related anxiety and fear 44 . Participants are asked to indicate the frequency they engage in various thoughts and activities in relation to their pain. The PASS is scored from 0 to 5 with 0 being “Never” and 5 being “Always”. Only adults completed the PASS measure, since no validated adolescent version exists currently. The PASS had a questionable reliability in our sample with a Cronbach’s α  = 0.638. Pain interference was measured using the Patient-Reported Outcomes Measurement Information System (PROMIS), which measures the extent to which pain hinders engagement with social, recreational, emotional, physical, and cognitive activities 45 . Pediatric participants completed the pediatric version of the PROMIS measure 46 . The scores are standardized to a mean = 50, standard deviation (SD) = 10 in a general US population. Standardized scores are categorized according to the PROMIS guidelines as: “Within normal limits”, T -score ≤ 55; “mild”, 55 <  T -score ≤ 60; “moderate”, 60 <  T -score ≥ 70; “severe”, 70 <  T -score ≥ 80. Higher scores reflect greater hindrance to engage in social, recreational, and other activities due to pain. For the pediatric pain interference measure, there are slightly different categorizations. The guidelines are as follows: “Within normal limits”, T -score ≤ 50; “mild”, 50 <  T -score ≤ 55; “moderate”, 55  65. Pain Interference measure had acceptable reliability in our sample (Cronbach’s α  = 0.706) All participants completed standardized measures of anxiety and depression symptoms. The PROMIS Depression and Anxiety measures, both 8-item inventories, assess depressive 47 , 48 or anxious mood 47 . Participants are asked to rate how frequently they experience certain symptoms over the last 7 days for both measures. Pediatric participants completed adapted measures for depression and anxiety that had the same standardized score scale 49 . The total score assesses both the range of experienced symptoms and the frequency of their occurrence. The scores are standardized to a mean = 50, SD = 10 in a general US population. Higher scores indicate greater anxious or depressive symptoms. Standardized scores are categorized according to Health Measures score cut-off points. For the adult surveys, standardized scores are categorized according to PROMIS guidelines: within normal limits, T -score ≤ 55; mild, 55 <  T -score ≤ 60; moderate, 60 <  T -score ≥ 70; severe, 70 <  T -score ≥ 80. For the pediatric measures, there are slightly different categorizations. The guidelines are as follows: “Within normal limits”, T -score ≤ 50; “mild”, 50 <  T -score ≤ 55; “moderate”, 55  65. Images were acquired using a 3 T Siemens Prisma scanner equipped with a 32-channel phased-array head coil. High resolution T1-weighted magnetization-prepared rapid acquisition gradient echo (MPRAGE) anatomical images were obtained from all participants with the following sequence parameters: repetition time (TR) = 2400 ms, echo time (TE) = 2.19 ms, field of view = 240 mm, flip angle = 8°, voxel size = 0.8 × 0.8 × 0.8 mm3; slice thickness = 0.8 mm; GRAPPA acceleration factor = 2. Structural images were analyzed to obtain measures for cortical thickness and cortical volume using standard processing pipelines in FreeSurfer software (version 7.1; https://surfer.nmr.mgh.harvard.edu ) 50 . Image processing included skull stripping, automated Talairach transformation, segmentation of subcortical gray and white matter structures, intensity normalization, tessellation of GM/white matter boundary, automated topology correction, and surface deformation. The technical details are described in prior publications 51 – 64 . Image outputs from FreeSurfer processing were visually inspected for inaccuracies before data being included for further analyses. Cortical output maps were spatially smoothed by applying a Gaussian kernel of 10 mm full width at half maximum. Statistical analysis for the self-report measures was analyzed using SPSS version 28.0.1.0. Descriptive statistics were calculated to determine mean, SD, minimum and maximum, and number of participants. Median and inter-quartile range were provided when the data were non-normal. Group differences were analyzed using chi-square tests for categorical values, independent sample t-tests for continuous and parametric values, and Mann–Whitney U Tests for continuous and non-parametric values. To evaluate differences of racial composition between the two groups, all races/ethnicities except for “White” were collapsed under a “Non-White” to run a chi-square test with a sufficient sample size to converge. Unless otherwise stated, mean ( M ) and SD are reported. A p value of 0.05 was used to determine statistical significance. To identify group differences in cortical thickness and cortical volume between participants with SCE and participants with NDE, and to evaluate the effect of age on these cortical measures, a general linear model was implemented at each vertex in the whole brain by using the “mri_glmfit” tool in FreeSurfer. Using our brain outcomes (cortical thickness/volume) as the dependent variable, we used age and group as our independent variables. Our primary research interests were to first explore an interaction between age and group. We aimed to evaluate if the relationship between age and cortical thickness was different between our SCE and NDE exposure groups. If significant, parameter estimates were extracted from the significant cluster and for each exposure group, we performed a correlation analysis between age and cortical thickness. Secondary research interests were focused on understanding group differences in brain metrics (cortical thickness/brain volume). We incorporated the variable estimated total intracranial volume (eTIV) and age as control variables when volume metrics were analyzed. All analyses were conducted for the right and left hemispheres separately. The resulting maps were cluster-wise corrected for multiple comparisons with Monte Carlo simulations (“mri_glmfit-sim” function, corrected for two hemispheres) with a threshold of p  < 0.05. Significant clusters were visualized on semi-inflated brain surfaces. We performed a quality assessment of output from Freesurfer using the qatools program. Output is in Supplementary Table  1 . Significant clusters from the group comparison analysis were binarized and extracted for each participant producing two variables for each participant (NDE and SCE). Extracted behavioral data reflected a reduced battery of tests based on the availability of data points for each participant (some participants had missing data based on factors such as not having time during the study visit). The included set of behavioral features contained: age, anxiety, depression, PASS, pain interference, and the PSQ. All data were standardized using custom-made scripts in Python. We performed affinity-based clustering to integrate both imaging and behavioral data for the group that included both NDE and SCE participants. Output clusters were displayed as a function of both brain volume metrics. Clusters were disaggregated and presented using radar plots to understand sub-group phenotypes. We elected to use two example clusters that showed prominent differences in brain volume metrics to understand within group correlation parameters. Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

Demographic and clinical characteristics are included in Table  1 . The total study population included 43 participants with SCE (13–43 years old, M age = 27.1 ± 8.4) and 26 participants with NDE (13—41 years old, M age = 25.5 ± 6.5). Education of participants with NDE ranged from currently in high school to having earned a Master’s Degree. Participants in the SCE group had educational backgrounds that ranged from currently in high school to having earned a PhD. A chi-squared test found no group differences for education ( X 2 (5) = 5.7, p  = 0.34). Table 1 Demographics of patients with surgically confirmed endometriosis and controls with no history of endometriosis Measure (SCE/NHE) Surgically confirmed endometriosis (SCE) No history of endometriosis (NHE) Test Statistic P value Demographics Age, years 27.1 ± 8.4 (13–43) 25.5 ± 6.5 (13–41) t  = −0.89 p  = 0.38 Gender, n (%) --- --- Female 41 (95) 26 (100) Non-binary 2 (5) 0 (0) Race, n (%) χ 2  = 3.13 p  = 0.07 White 35 (81) 16 (62) Non-White 8 (19) 10 (38) Black 3 (7) 1 (4) Asian 2 (5) 5 (19) Other 1 (2) 4 (15) Multiple 2 (5) 0 (0) Education level, n (%) --- --- In HS 6 (14) 1 (4) HS 11 (26) 8 (31) Associates degree 2 (5) 0 (0) Bachelors degree 20 (47) 12 (46.2) Masters degree 3 (7) 5 (19) PhD 1 (2) 0 (0) Menstrual cycle variables Age of menarche, years ( n  = 41/26) Med: 12 (Q: 11,13) Med: 12 (Q: 11, 13.25) U  = 485.5 p  = 0.53 Days on period ( n  = 35/20) Med: 7 (Q: 5, 8) Med: 5 (Q: 4, 5) U  = 131 p  < 0.001 Pain level during menstrual cycle, (0–10) ( n  = 43/25) Med: 7 (Q: 6, 8) Med: 3 (Q: 1.5, 5) U  = 150 p  < 0.001 Pelvic pain days per month ( n  = 32) 7.9 ± 9.6 (0–30) --- --- --- Currently using hormonal medication, n (%) ( n  = 43/26) 32 (74) 16 (62) --- --- Menstrual suppression (hormonal or surgical), n (%) ( n  = 43/26) 19 (44) 3 (12) --- --- Age of suppression, years --- --- Under 20 10 (67) 0 (0) 20–29 2 (13) 3 (100) 30–39 3 (30) 0 (0) 40–49 0 (0) 0 (0) Data are expressed as mean ± SD or number of patients (%). Continuous data were analyzed using independent sample t-test for parametric values and Mann–Whitney U test for non-parametric values. Categorical data was analyzed using a chi-square (χ 2 ) test. Race group differences were analyzed using a chi-square test comparing those reporting White and other race/ethnicity groups combined due to sample size limitations. HS high school. The p values that reached the significance threshold are bolded ( p  < 0.05). Participant numbers for each scale are reported in parentheses as: ( n  = surgically confirmed endometriosis/no history of endometriosis). Med median, Q quartile (25%, 75%). Demographics of patients with surgically confirmed endometriosis and controls with no history of endometriosis 27.1 ± 8.4 (13–43) 25.5 ± 6.5 (13–41) Med: 12 (Q: 11,13) Med: 12 (Q: 11, 13.25) Med: 7 (Q: 5, 8) Med: 5 (Q: 4, 5) Med: 7 (Q: 6, 8) Med: 3 (Q: 1.5, 5) 7.9 ± 9.6 (0–30) Data are expressed as mean ± SD or number of patients (%). Continuous data were analyzed using independent sample t-test for parametric values and Mann–Whitney U test for non-parametric values. Categorical data was analyzed using a chi-square (χ 2 ) test. Race group differences were analyzed using a chi-square test comparing those reporting White and other race/ethnicity groups combined due to sample size limitations. HS high school. The p values that reached the significance threshold are bolded ( p  < 0.05). Participant numbers for each scale are reported in parentheses as: ( n  = surgically confirmed endometriosis/no history of endometriosis). Med median, Q quartile (25%, 75%). The mean age of menarche was only four months younger for participants with SCE compared to participants with NDE. The SCE group reported a longer duration of menstrual bleeding. The SCE group also had greater menstrual pain (rated from 0 to 10) compared to the NDE group, by design as participants with chronic pain were excluded from the NDE group. More individuals with SCE used hormonal medication compared to participants with NDE. All medication was hormonal contraception except for one patient who uses Orilissa, a GnRH antagonist oral medication that is used to treat hormonal contraception-resistant endometriosis-associated pain 65 . For those who reported being on contraceptives, estradiol, progesterone, progestin/estrogen formulations were used. A greater number of participants with SCE reported menstrual suppression or stopping of menstrual periods due to hormonal or surgical influences compared to the NDE group. For endometriosis-specific clinical characteristics, see Table  2 . As rASRM stage documentation is not standard of care at the participating clinical sites, 42% of participants in the SCE group were missing stage at the most recent surgery. As shown in Table  2 , these participants had stages that ranged from 1 to 4, which correlates with lesion volume and consequently surgical complexity, but not with pain symptom severity nor pain-focused treatment prognosis. On the day of the visit, the SCE group endorsed a greater current pain score compared to the NDE group. Table 2 Clinical endometriosis variables and psychological variables for patients with surgically confirmed endometriosis and controls with no history of endometriosis Measure Surgically confirmed endometriosis (SCE) No history of endometriosis (NHE) Test statistic P value Endometriosis clinical variables  Years since diagnosis ( n  = 43) 3.5 ± 3.8 (0–14) --- --- ---  Surgery # for endometriosis ( n  = 43) 1.2 ± 0.8 (0–4) --- --- ---  Endometriosis stage ( n  = 25) 1.6 ± 1.2 (1–4) --- --- ---  Endometriosis impact ( n  = 43) 42.5 ± 20.4 (2.4–83.9) --- --- --- Pain experience  Current pain score (at visit), 0–10 ( n  = 40/23) 1.9 ± 1.9 (0–7) 0.2 ± 0.6 (0–2) --- ---  Pain Catastrophizing ( n  = 43/26) Med: 21 (Q: 11, 28) Med: 2 (Q: 0, 14.25) U  = 222 p  <  0.001  Pain anxiety symptom scale ( n  = 37/25) 26.1 ± 18.8 (3–83) 29.8 ± 20.0 (0–70) t  = 3.25 p  =  0.002  Pain sensitivity questionnaire ( n  = 43/26) 3.4 ± 1.3 (0.9–5.6) 3.8 ± 1.3 (1.5–6) t  = 1.09 p  = 0.28  Pain interference, T -Score ( n  = 42/26) Med: 54.85 (Q: 40.7, 62.1) Med: 40.7 (Q: 40.7, 53.85) U  = 354 p  =  0.01  Promise depression, T -Score ( n  = 42/26) 52.7 ± 9.5 (35.2–72.8) 51.1 ± 9.0 (38.2–72.6) t  = 0.68 p  = 0.50  Promise anxiety, T -Score ( n  = 42/26) 56.8 ± 8.3 (37.1–73.0) 54.7 ± 10.0 (37.1–74.6) t  = 0.91 p  = 0.37 Statistically significant p  < 0.05 values are in bold. Data are expressed as mean ± SD or number of patients (%). Continuous data was analyzed using independent sample t-test for parametric values and Mann–Whitney U test for non-parametric values. Significant p values are bolded ( p  < 0.05). Med  median, Q  quartile (25%, 75%). Clinical endometriosis variables and psychological variables for patients with surgically confirmed endometriosis and controls with no history of endometriosis 3.5 ± 3.8 (0–14) 1.2 ± 0.8 (0–4) 1.6 ± 1.2 (1–4) 42.5 ± 20.4 (2.4–83.9) 1.9 ± 1.9 (0–7) 0.2 ± 0.6 (0–2) Med: 21 (Q: 11, 28) Med: 2 (Q: 0, 14.25) 26.1 ± 18.8 (3–83) 29.8 ± 20.0 (0–70) 3.4 ± 1.3 (0.9–5.6) 3.8 ± 1.3 (1.5–6) Med: 54.85 (Q: 40.7, 62.1) Med: 40.7 (Q: 40.7, 53.85) 52.7 ± 9.5 (35.2–72.8) 51.1 ± 9.0 (38.2–72.6) 56.8 ± 8.3 (37.1–73.0) 54.7 ± 10.0 (37.1–74.6) Statistically significant p  < 0.05 values are in bold. Data are expressed as mean ± SD or number of patients (%). Continuous data was analyzed using independent sample t-test for parametric values and Mann–Whitney U test for non-parametric values. Significant p values are bolded ( p  < 0.05). Med  median, Q  quartile (25%, 75%). Pain behaviors and emotional functioning metrics are included in Table  2 along with corresponding statistical findings. The groups had less than two points difference in mean anxiety and depression symptoms scores (both p values p  > 0.37). The SCE group reported significantly more pain-related anxiety symptoms, as well as greater pain catastrophizing and pain interference compared to the NDE group ( p  < 0.01). There was no difference in mean pain sensitivity scores between groups ( p  = 0.28). An interaction between age and cortical thickness was found only in the right hemisphere in the superior frontal gyrus (cluster max = 3.99, cluster size = 938.59 mm 2 , location: X  = 10.8, Y  = 57.4, Z  = 2.5, p  = 0.02). As shown in Fig.  1 , whereas the NDE group had no significant relationship between age and cortical thickness ( r (26) = −0.17, p  = 0.40), persons in the SCE group had a significant negative slope ( r (43) = −0.62, p  ≤ 0.001). Exploratory correlation analysis dividing our groups into younger and older (break point = 25 years), showed that only in the SCE-young group was there a significant negative relationship ( r (18) = −0.75, p  < 0.001); all other relationships were not significant (SCE older: r (21) = 0.25, p  = 0.24; NDE Young: r (11) = −0.17, p  = 0.57); NDE Older: r (11) = 0.40, p  = 0.17. There was no interaction observed in the left hemisphere ( p  > 0.05) and there were no differences between groups when controlling for eTIV and age ( p  > 0.05). Fig. 1 Plots of significant clusters (i) and scatter plots showing individual data from significant clusters (ii). The relationship between cortical thickness of the superior frontal gyrus and age in both the control and SCE groups is plotted. A regression line is added for the SCE group to demonstrate the significant relationship between features that was not present in the NDE group. The relationship between cortical thickness of the superior frontal gyrus and age in both the control and SCE groups is plotted. A regression line is added for the SCE group to demonstrate the significant relationship between features that was not present in the NDE group. Cortical thickness from the superior frontal gyrus was correlated against Pain Interference, Average pain during menstrual cycle, Pain on day of study visit and the PSQ. No significant correlations were found in the SCE cohort (ps > 0.23). Only Pain interference was significantly correlated in the NHE cohort ( r (26) = −0.45, p  = 0.02); all other relationships were non-significant (ps > 0.23). There was no interaction between SCE versus NDE group and age to predict volume in either the left or right hemisphere ( p  > 0.05). In the right hemisphere (Fig.  2 ), we greater volume in the lateral occipital region (cluster max = 2.31, size = 943.74 mm 2 , location: X  = 28.5, Y  = −95.1, Z  = −9.6) after controlling for age and ETIV in the NDE group ( M volume  = 2.16, SD = 0.28) relative to the SCE ( M  = 2.11, SD = 0.35) group ( p  = 0.02; Cohen’s D  = 0.37). In the left hemisphere, we observed greater volume in the fusiform gyrus (cluster max = 2.89, size = 891.56 mm 2 , location: X −37.8, −44.9, −21.8) in the NDE group (M volume  = 2.51, SD = 0.27) relative to the SCE ( M  = 2.5, SD = 0.39) group ( p  = 0.03; Cohen’s D  = 0.33). Fig. 2 Group differences in brain volume. Significant differences are highlighted in blue in the left and right hemisphere. Two regions of significance were observed that are highlighted. P values significant at p  < 0.05. Significant differences are highlighted in blue in the left and right hemisphere. Two regions of significance were observed that are highlighted. P values significant at p  < 0.05. We explored our volumetric findings using affinity-based clustering to understand the relationship between volumetric differences and behavioral reporting in our entire study population (NDE and SCE). As shown in Fig.  3 where the brain volumes ( z -scores) are presented on respective x - and y -axes, there is a symmetrical relationship between volume of these two regions. The overlayed cluster distribution—a composite of brain imaging and clinical/demographic features—demonstrates that the eight observed clusters have unique properties relative to the brain volumetric data (e.g., see Cluster 1 versus Cluster 8). Plotting each feature in radar plots and heat maps (Fig.  4 ) demonstrates the unique properties of each feature domain and cluster, respectively. To further explore these relationships, we elected to examine Cluster 1 and 2 at the single subject level. Cluster 1 includes 2 control participants and 8 SCE participants, whilst Cluster 2 includes 4 participants with SCE only. As can be seen in Fig.  5 , there is a clear difference in brain volumes between groups for both the fusiform gyrus and lateral occipital cortex. In terms of behavior, Cluster 1 appeared to express higher levels of depression, anxiety, and pain interference. Alternatively, for Cluster 2, depression, anxiety, and pain interference were lower, despite high levels of reporting on the PASS. For an integrated picture, a correlation heat map was performed for these two clusters, which showed for both that the two brain regions tended to be correlated with each other, and that for each cluster, there is a unique brain-behavior relationship (as indicated by differences in shading in the correlation matrix heat map). Fig. 3 Affinity propagation cluster analysis output. Input features included age, anxiety and depression self-reporting, pain specific behaviors, and brain imaging. Brain volume ( z -score) from the fusiform gyrus and the lateral occipital cortex are shown on the x - and y -axis, respectively. The eight observed clusters are plotted in different randomized colors. Data included in this cluster analysis is from clinical and control groups. Fig. 4 Radar plots outlining the distribution of feature representation amongst the produced clusters (1 through 8). Mean z -score values are plotted in each radar plot that includes data from both the NDE and SCE groups. Fig. 5 An enhanced view of Clusters 1 and 2 as exemplar figures. The y -axis (Score) reflects a general scale as all variables are presented in their native units (age = years; anxiety, depression, pain interference =  T -score; PSQ = raw total score). Comparing the two clusters side by side, Cluster 1 has reduced volume (Left) relative to Cluster 2 (Right); however, there are unique psychological components associated with each cluster group. That is, Cluster 1 is more marked by elevated levels of anxiety depression and pain-interference, whereas Cluster 2 is more marked by elevated PASS scores. Lastly, side-by-side heat maps suggest a different relationship between brain changes and psychological scores. Green circles around participant numbers indicate a healthy control. Input features included age, anxiety and depression self-reporting, pain specific behaviors, and brain imaging. Brain volume ( z -score) from the fusiform gyrus and the lateral occipital cortex are shown on the x - and y -axis, respectively. The eight observed clusters are plotted in different randomized colors. Data included in this cluster analysis is from clinical and control groups. Mean z -score values are plotted in each radar plot that includes data from both the NDE and SCE groups. The y -axis (Score) reflects a general scale as all variables are presented in their native units (age = years; anxiety, depression, pain interference =  T -score; PSQ = raw total score). Comparing the two clusters side by side, Cluster 1 has reduced volume (Left) relative to Cluster 2 (Right); however, there are unique psychological components associated with each cluster group. That is, Cluster 1 is more marked by elevated levels of anxiety depression and pain-interference, whereas Cluster 2 is more marked by elevated PASS scores. Lastly, side-by-side heat maps suggest a different relationship between brain changes and psychological scores. Green circles around participant numbers indicate a healthy control.

Our findings align with and extend prior work on persons with endometriosis-associated pain and provide evidence to suggest there are quantifiable age-related brain metrics that differ between those with and without endometriosis. Clinically, what should be appreciate that is that central brain structural changes appear in persons with SCE, and that such changes may have an onset in the pediatric setting. We provide exploratory findings to suggest that regional brain changes may confer unique patient phenotypes. Future research will be required to replicate these findings and expand the population studied to include individuals in the pre- and post-menopausal state and in longitudinal analyses.

Study findings demonstrate support for age-related correlations in brain structure that are unique to persons with surgically confirmed endometriosis. Persons with SCE were found to exhibit lower relative volume in the lateral occipital cortex and fusiform gyrus, as well as a negative relationship between age and cortical thickness that was unique to the SCE group in the superior frontal cortex. Exploratory analyses provide insight into sub-groups with unique brain-behavior relationships and further our understanding regarding the cross-sectional relationships between endometriosis-associated pain and the brain. Participants reported a diverse range of self-reported symptoms and experiences associated with their diagnosis. For those with endometriosis, on average, the time since diagnosis was 3.5 ± 3.8 years. This time frame should be viewed in light of the well-documented high rates of delayed diagnosis in adult women 7 , 8 , 31 suggesting the length of time this group experienced pain may extend years before this reported period. Those with SCE reported minimal to severe impact of the disease on their lives, which underscores the heterogeneous impacts of endometriosis on a person’s daily life. On average, participants with SCE reported a moderate impact of endometriosis on the EIQ (average impact score of 42.50). Participants with SCE also reported significantly more days of bleeding and higher intensity of menstrual pain relative to the NDE group, in line with prior literature on risk factors for endometriosis 1 , 66 . There was no difference between hormonal contraceptive use between groups. We tried to control hormonal levels as much as possible by limiting testing of natural cycling participants to days 2–10. This was to make hormonal levels as consistent as possible to match all participants in their follicular phase with relatively low levels of progesterone and estrogen. As such, we believe our group captures the clinical variability observed by persons diagnosed with endometriosis. In the present study, two exposure groups were compared to observe brain-related changes cross-sectionally associated with endometriosis-associated pain. As such, persons in the NDE group were eligible only if they reported no evidence of current or chronic pain. As this control measure precludes accurate group-level comparisons of pain metrics, we evaluated pain-behaviors as trait, rather than state, functions. Participants with SCE showed elevated pain catastrophizing, pain anxiety, and pain interference compared to controls. Several studies have found that individuals with endometriosis are more likely to engage in pain catastrophizing, which can lead to hypervigilance and enhanced sensitivity to pain 19 , 22 . Pain catastrophizing, defined as magnified worry and negative thinking about the pain experience, is also related to more severe CPP, which is a frequent symptom of endometriosis 20 . Pain-related cognitive and affective factors are known to mediate the chronic pain experience and are associated with several brain regions involved in pain processing, such as attention to pain, pain perception, and top-down pain inhibition 67 – 71 . Pain-related cognitive and affective components are also linked to depression, pain-related disability, and treatment outcomes 72 – 76 . Together, findings align with prior literature in individuals with endometriosis-associated pain and demonstrate the diffuse impact of pain on behavior and daily living. Current findings add to the growing literature demonstrating central brain differences in persons with pelvic pain. Prior work in persons with CPP and endometriosis has shown brain regions such as thalamus, cingulate, putamen and insula to have smaller volume 27 , whereas research in persons with dysmenorrhea has shown changes in regions such as the ACC, and superior temporal gyrus to have positive correlations between GM volume and pain severity scores (MPQ and pain duration, respectively) 77 . Findings from the current study reflect changes to both volume and thickness metrics for those with, compared to those without, endometriosis. A significant decrease in cortical thickness in the right superior frontal gyrus was found to be associated with older age in persons with SCE. Prior brain-development research has largely focused on persons roughly 20 years onward, finding a relative plateau in frontal lobe cortical thickness in individuals aged 20–50 years old 78 . Some studies have found increasing brain volume in frontal regions from 20 to 50 years old 79 , and others have shown there is a relative increase until 35 years old 80 , and a decrease roughly 35 onwards 79 , 80 . Notably, brain development has significant inter-individual and inter-regional variation, particularly in pediatrics 81 . As such, we cannot directly evaluate our findings, which reflect a study-observed region of interest, to those from larger datasets, which typically define regions through existing atlases. Our findings suggest that in persons with SCE, there is a negative correlation between cortical thickness and age (13–43 years old) in a sub-region of the superior frontal gyrus. Exploratory analyses suggest this was primarily driven by those in the younger (<25 years) group. Although there is not much published research on this subject, findings may align with prior work focused on pre-adolescence to adolescence, showing a peak in frontal lobe volumes in females around the age of 11, with declines thereafter 82 . This would align with both cohorts showing negative correlations (not significant in NHE). As to why this relationship is more pronounced in the SCE cohort, it is notable that cortical thickness did not correlate with any pain metric in our SCE cohort. As such, findings likely do not pertain to the direct impact of pain on cortical health. However, in a group of persons with dysmenorrhea, Tu and colleagues observed changes in the volume of the superior frontal gyrus to negatively correlate with recalled pain severity scores 77 . This brain region has been implicated in higher-order cognitive functions, including learning and memory, attention, problem-solving, decision making, as well as emotional processing 83 – 88 . As it relates to pain, this region may coincide with work showing that decreased coupling between the primary sensory region and the medial prefrontal cortex was associated with an increase in self-reported pain after surgical procedures 89 . Similar work showing connectivity between the medial prefrontal cortex and the nucleus accumbens predicted chronic pain status 90 would align with a role for this region in pain modulation, rather than pain detection. This interpretation would be in line with chronic pain populations showing decreased glutamate concentrations and abnormal functional connectivity in the medial superior frontal gyrus 85 , 86 , as well as decreased activity in pain exposure tasks 91 . As such, findings may reflect group-level differences in pain modulation. Group differences in brain volume were observed in well-controlled analyses. When controlling for age and eTIV, we were able to observe a lower volume in the SCE relative to the healthy participant group in two brain regions. As it pertains to the fusiform gyrus, prior work in persons with fibromyalgia has demonstrated its activation to be associated with pain perception and the cognitive appraisal of pain 92 . In persons with chronic back pain, fMRI activity within the fusiform gyrus (left hemisphere) was found to be implicated in encoding pain intensity 93 . Interestingly, the lateral occipital cortex has also been observed to encode pain intensity in a person with chronic back pain, drawing parallels between these two brain region findings 91 . Other researchers have found activity in this region (associated with visual object recognition) to be modulated by the presence of pain 94 . This could perhaps align with observations that activity can be decreased in this region in the context of pain 95 , and the smaller volume of this region in our study population, alluding to a form of potential atrophy due to prolonged pain exposure. In the current data set, we are limited in our ability to resolve function-behavior relationships without fMRI; however, it is notable that lower brain volume relationships occurred alongside elevated reporting on pain catastrophizing, pain anxiety symptoms, pain interference, and elevated pain during the menstrual cycle. Future research using fMRI will be required to resolve this association. To better understand the significance of brain volume changes, we elected to explore our findings using clustering analysis. We used affinity-based clustering (see prior uses of this technique in neuroimaging: Li and Horvath 96 , Liu et al. 97 , Zhang et al. 98 , Duan and Zhou 99 ) as it internally defines the optimal number of intrinsic clusters while integrating multi-variable dynamics. Therefore, each cluster is a reflection not of one parameter but of all parameters that we submitted to the algorithm. We integrated both NDE and SCE participants and chose a reduced set of metrics to integrate alongside brain volume changes based on their role in pain and affective health, alongside the availability of data points across participants. As shown in Fig.  2 , there is evidence to suggest that there may be embedded relationships in our data. Despite a seemingly linear relationship between changes in the two observed brain regions, there are sub-groups (see color distribution) where this relationship appears to fluctuate (see Parise et al. 93 ; Mayr et al. 91 ). This is demonstrated in the radar plots and heat maps of Fig.  3 . On the left side, it can be appreciated that features have a different representation between clusters expanding on our findings across groups. The heat maps provide added depth to this analysis and show how features are interrelated within each group. Across clusters, the heat maps take on very different representations and demonstrate, for example, that the relationship between brain volume (from the two identified brain regions) and behavioral features is different across clusters. This is further demonstrated in the exemplar plot (see Fig.  4 ), where brain volume appears to be more highly correlated with behavioral features in Cluster 1 than in Cluster 2. Notably, a byproduct of affinity clustering analysis is that some groups are too small to evaluate in terms of correlation; however, such grouping is still considered unique in their expression of each variable, relative to the other clusters. These exploratory findings provide evidence to suggest that (1) pain behaviors may not have the same neurological basis in all persons with SCE, and (2) that sub-groups need to be identified to tailor behavioral versus neurological-focused interventions using multi-variable approaches. Some limitations of the present study should be mentioned. First, we investigated age-related brain differences with cross-sectional data. Prior studies have commented that cross-sectional approaches may underestimate brain changes relating to age 100 , with longitudinal analyses being required to understand intra-individual brain changes over time 101 . Second, we only evaluated thickness and volume metrics, and further studies are needed to explore age-related differences using multimodal neuroimaging data with additional structural, functional, and behavioral measures (i.e., using paradigms examining the pain modulation system specifically). Third, this study included a relatively small sample size. Prior investigations evaluating brain-age correlations have involved upwards of 100 healthy persons (see Giedd et al., for example) 82 . The current sample was the largest neuroimaging sample of adolescents with SCE to date, to the knowledge of the authors. As such, our findings should be justly interpreted. To fully understand the impact of this, and other gynecological conditions, both prior to and after menses, larger multi-modal studies are required for data modeling and to inform clinical intervention. Fourth, all participants in the SCE group had surgically diagnosed endometriosis, indicating that they reported symptoms severe enough to warrant surgery. This does not preclude the occurrence of endometriosis in the NDE group but rather limits our conclusions to pain in the context of endometriosis. As noted in Table  1 , although our healthy participant groups had exclusions relating to chronic pain, persons still reported active (>0) pain during normal menstrual function. We do not believe this interfered with current findings but rather supports the perspective that more information is needed in women’s health to better define and respect pain symptoms. Finally, we note that future work can integrate a more in-depth analysis of current medications, including hormonal and non-hormonal prescriptions, to better understand their role in pain reporting, as well as more precise education levels for data modeling and analyses.

Endometriosis is a prevalent and often painful gynecological condition where abnormal endometrial tissue grows outside the uterus 1 , 2 , and can be associated with chronic inflammation, chronic pain, and infertility 1 , 3 , 4 . Previous studies estimate a prevalence of 10% of biological females of reproductive age to have endometriosis 5 . Endometriosis has been diagnosed in people as young as eight years old 6 , although the average age at diagnosis remains in the early 30 s. Most with endometriosis experience symptoms for several years before successful diagnosis, with the average diagnosis occurring seven years after onset of symptoms 1 , 7 , 8 . Despite evidence of disease burden from adolescence through adulthood, no research to date has explored age-related correlations of this condition on brain health. The stages of brain development during early to late adolescents present a highly vulnerable time frame for pain exposure as neuronal networks evolve and consolidate. Early life exposure to pain from infancy through puberty has been associated with poor cognitive, physical, and psychological function later in life 9 – 13 . In adults, pain is well established to lead to cognitive (e.g., attention, perception, executive functioning) 14 , and mental health (e.g., depression 15 ) impairments 16 . Chronic pain, at all ages, is associated with changes in the central nervous system (CNS), particularly with stress reactivity and long-term functional and morphometric brain changes 13 , 17 . Persistent pain-related changes in autonomic arousal and cortisol reactivity disrupt stress reactivity and are associated with psychological problems later in life 13 . Pain behaviors, such as pain catastrophizing and pain sensitivity, as well as psychological factors like anxiety and depression, play a role in a person’s disease experience 18 – 22 . Structural changes in the CNS may perpetuate transitions to chronic pain, predict pain vulnerability  23 , and contribute to learning and memory deficits as well as negative cognitive and affective states 15 , 17 , 23 . Pain-related structural changes in regions such as the amygdala, anterior cingulate cortex, and prefrontal cortex modify many important cognitive and psychological domains, including pain processing, cognitive control, emotional regulation, and memory 13 , 17 . Accordingly, it is imperative to integrate pain behaviors into our understanding for healthy neurological development. Chronic pain has an established and diffuse impact on brain health. Prior work on pain matrices and connectomes underscores a diffuse impact of pain-responsive regions of the brain that include, for example, the insula 24 . A recent meta-analysis on neuroimaging studies in chronic pain corroborates this finding, with notable alterations in the insular cortex, superior temporal cortex and superior frontal gyrus using structural and resting state/task-based functional MRI 25 . The regional extent of pain sensitivity is a notable factor. In a cohort of all female subjects, the superior frontal gyrus was found to have decreased cortical thickness in persons with fibromyalgia (diffuse pain) relative to healthy controls, which was not observed in a cohort of persons with trigeminal neuralgia (regional pain). Moreover, cortical thickness of the superior frontal gyrus and volume of the right thalamus were found to negatively correlate with the extent of orofacial pain intensity  26 . Endometriosis patients with chronic pelvic pain (CPP) demonstrate smaller gray matter (GM) volume in cortical and sub-cortical regions, including the thalamus, cingulate, putamen and insula 27 . Functional connectivity analyses in persons with endometriosis and CPP have focused on the anterior insula as a seed region and shown decreased functional connectivity with the right cerebellum and left middle frontal gyrus 28 , 29 . Although there is evidence to suggest central brain changes in persons with endometriosis-associated pain, it is unclear if the age-related development of brain tissue is impacted in persons with endometriosis-associated pain. In this cross-sectional study, we examine the impact of endometriosis-associated pain on the brain in individuals recruited across the reproductive lifespan. We aimed to: (1) evaluate structural brain differences between groups with surgically confirmed endometriosis and persons with no diagnosis of endometriosis (pain-free); and (2) evaluate the relationship between age and cortical thickness/brain volume for an interaction based on group membership. We hypothesized that there would be unique age-related brain changes in persons with endometriosis and that observed brain changes would be related to unique participant phenotypes.

Supplementary Information Reporting Summary Supplementary Information Reporting Summary