https://doi.org/10.1093/braincomms/fcac020 BRAIN COMMUNICATIONS 2022: Page 1 of 12 | 1 BRAIN AIN COMMUNICATIONS Sex-specific lesion pattern of functional outcomes after stroke Anna K. Bonkhoff,1 Martin Bretzner,1,2 Sungmin Hong,1 Markus D. Schirmer,1,3 Alexander Cohen,4,5 Robert W. Regenhardt,1 Kathleen L. Donahue,1 Marco J. Nardin,1 Adrian V. Dalca,6,7 Anne-Katrin Giese,8 Mark R. Etherton,1 Brandon L. Hancock,7 Steven J. T. Mocking,7 Elissa C. McIntosh,7 John Attia,9,10 Oscar R. Benavente,11 Stephen Bevan,12 John W. Cole,13 Amanda Donatti,14 Christoph J. Griessenauer,15,16 Laura Heitsch,17,18 Lukas Holmegaard,19,20 Katarina Jood,19,20 Jordi Jimenez-Conde,21 Steven J. Kittner,13 Robin Lemmens,22,23 Christopher R. Levi,10,24 Caitrin W. McDonough,25 James F. Meschia,26 Chia-Ling Phuah,18 Arndt Rolfs,27 Stefan Ropele,28 Jonathan Rosand,1,7,29 Jaume Roquer,21 Tatjana Rundek,30 Ralph L. Sacco,30 Reinhold Schmidt,28 Pankaj Sharma,31,32 Agnieszka Slowik,33 Martin Söderholm,34,35 Alessandro Sousa,14 Tara M. Stanne,36 Daniel Strbian,37 Turgut Tatlisumak,20,38 Vincent Thijs,39,40 Achala Vagal,41 Johan Wasselius,42,43 Daniel Woo,44 Ramin Zand,45 Patrick F. McArdle,46 Bradford B. Worrall,47,48 Christina Jern,36,49 Arne G. Lindgren,50,51 Jane Maguire,52 Michael D. Fox,4 Danilo Bzdok,53,54 Ona Wu5 and Natalia S. Rost1on behalf of the MRI-GENIE and GISCOME Investigators and the International Stroke Genetics Consortium Stroke represents a considerable burden of disease for both men and women. However, a growing body of literature suggests clinically relevant sex differences in the underlying causes, presentations and outcomes of acute ischaemic stroke. In a recent study, we reported sex divergences in lesion topographies: specific to women, acute stroke severity was linked to lesions in the left-hemispheric posterior circulation. We here determined whether these sex-specific brain manifestations also affect long-term outcomes. We relied on 822 acute ischaemic patients [age: 64.7 (15.0) years, 39% women] originating from the multi-centre MRI-GENIE study to model un- favourable outcomes (modified Rankin Scale .2) based on acute neuroimaging data in a Bayesian hierarchical framework. Lesions encompassing bilateral subcortical nuclei and left-lateralized regions in proximity to the insula explained outcomes across men and women (area under the curve = 0.81). A pattern of left-hemispheric posterior circulation brain regions, combining left hippo- campus, precuneus, fusiform and lingual gyrus, occipital pole and latero-occipital cortex, showed a substantially higher relevance in explaining functional outcomes in women compared to men [mean difference of Bayesian posterior distributions (men – women) = −0.295 (90% highest posterior density interval = −0.556 to −0.068)]. Once validated in prospective studies, our findings may mo- tivate a sex-specific approach to clinical stroke management and hold the promise of enhancing outcomes on a population level. 1 J. Philip Kistler Stroke Research Center, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA 2 Univ. Lille, Inserm, CHU Lille, U1171—LilNCog (JPARC)—Lille Neurosciences & Cognition, Lille F-59000, France 3 Clinic for Neuroradiology, University Hospital Bonn, Bonn, Germany 4 Center for Brain Circuit Therapeutics, Department of Neurology, Psychiatry, and Radiology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA 5 Department of Neurology, Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA 6 Computer Science and Artificial Intelligence Lab, Massachusetts Institute of Technology, Boston, MA, USA Received October 04, 2021. Revised November 30, 2021. Accepted January 31, 2022. Advance access publication February 2, 2022 © The Author(s) 2022. Published by Oxford University Press on behalf of the Guarantors of Brain. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 2 | BRAIN COMMUNICATIONS 2022: Page 2 of 12 A. K. Bonkhoff et al. 7 Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, MA, USA 8 Department of Neurology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany 9 Hunter Medical Research Institute, Newcastle, NSW, Australia 10 School of Medicine and Public Health, University of Newcastle, Newcastle, NSW, Australia 11 Department of Medicine, Division of Neurology, University of British Columbia, Vancouver, BC, Canada 12 School of Life Sciences, University of Lincoln, Lincoln, UK 13 Department of Neurology, University of Maryland School of Medicine and Veterans Affairs Maryland Health Care System, Baltimore, MD, USA 14 School of Medical Sciences, University of Campinas (UNICAMP) and the Brazilian Institute of Neuroscience and Neurotechnology (BRAINN), Campinas, SP, Brazil 15 Department of Neurosurgery, Geisinger, Danville, PA, USA 16 Research Institute of Neurointervention, Paracelsus Medical University, Salzburg, Austria 17 Department of Emergency Medicine, Washington University School of Medicine, St Louis, MO, USA 18 Department of Neurology, Washington University School of Medicine & Barnes-Jewish Hospital, St Louis, MO, USA 19 Department of Clinical Neuroscience, Institute of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden 20 Department of Neurology, Sahlgrenska University Hospital, Gothenburg, Sweden 21 Department of Neurology, Neurovascular Research Group (NEUVAS), IMIM-Hospital del Mar (Institut Hospital del Mar d’Investigacions Mèdiques), Universitat Autonoma de Barcelona, Barcelona, Spain 22 Department of Neurosciences, Experimental Neurology and Leuven Research Institute for Neuroscience and Disease (LIND), KU Leuven—University of Leuven, Leuven, Belgium 23 Department of Neurology, VIB, Vesalius Research Center, Laboratory of Neurobiology, University Hospitals Leuven, Leuven, Belgium 24 Department of Neurology, John Hunter Hospital, Newcastle, NSW, Australia 25 Department of Pharmacotherapy and Translational Research and Center for Pharmacogenomics, University of Florida, Gainesville, FL, USA 26 Department of Neurology, Mayo Clinic, Jacksonville, FL, USA 27 Centogene AG, Rostock, Germany 28 Department of Neurology, Clinical Division of Neurogeriatrics, Medical University Graz, Graz, Austria 29 Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA, USA 30 Department of Neurology and Evelyn F. McKnight Brain Institute, Miller School of Medicine, University of Miami, Miami, FL, USA 31 Institute of Cardiovascular Research Royal Holloway, University of London (ICR2UL), London, UK 32 St Peter’s and Ashford Hospital, Egham, UK 33 Department of Neurology, Jagiellonian University Medical College, Krakow, Poland 34 Department of Clinical Sciences Malmö, Lund University, Lund, Sweden 35 Department of Neurology, Skåne University Hospital, Lund and Malmö, Malmo, Sweden 36 Department of Laboratory Medicine, Institute of Biomedicine, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden 37 Department of Neurology, Helsinki University Hospital and University of Helsinki, Helsinki, Finland 38 Department of Clinical Neuroscience, Institute of Neuroscience and Physiology, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden 39 Stroke Division, Florey Institute of Neuroscience and Mental Health, Heidelberg, Australia 40 Department of Neurology, Austin Health, Heidelberg, Australia 41 Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, OH, USA 42 Department of Clinical Sciences Lund, Radiology, Lund University, Lund, Sweden 43 Department of Radiology, Neuroradiology, Skåne University Hospital, Lund, Sweden 44 Department of Neurology and Rehabilitation Medicine, University of Cincinnati College of Medicine, Cincinnati, OH, USA 45 Department of Neurology, Geisinger, Danville, PA, USA 46 Division of Endocrinology, Diabetes and Nutrition, Department of Medicine, University of Maryland School of Medicine, Baltimore, MD, USA 47 Department of Neurology, University of Virginia, Charlottesville, VA, USA 48 Department of Public Health Sciences, University of Virginia, Charlottesville, VA, USA 49 Department of Clinical Genetics and Genomics, Sahlgrenska University Hospital, Gothenburg, Sweden 50 Department of Neurology, Skåne University Hospital, Lund, Sweden 51 Department of Clinical Sciences Lund, Neurology, Lund University, Lund, Sweden 52 University of Technology Sydney, Sydney, Australia 53 Department of Biomedical Engineering, McConnell Brain Imaging Centre, Montreal Neurological Institute, Faculty of Medicine, School of Computer Science, McGill University, Montreal, Canada 54 Mila—Quebec Artificial Intelligence Institute, Montreal, Canada Sex-specific lesion pattern BRAIN COMMUNICATIONS 2022: Page 3 of 12 | 3 Correspondence to: Anna K. Bonkhoff, J. Philip Kistler Stroke Research Center Massachusetts General Hospital, Harvard Medical School 175 Cambridge St, Suite 300 Boston, MA 02114, USA E-mail: abonkhoff@mgh.harvard.edu Keywords: acute ischaemic stroke; functional outcomes; sex differences; lesion patterns; Bayesian hierarchical modelling Abbreviations: AIS = acute ischaemic stroke; AUC = area under the receiver operating characteristic curve; DWI = diffusion-weighted imaging; HO = Harvard–Oxford; HPDI = highest probability density interval; LNM = lesion network mapping; mRS = modified Rankin Scale; NMF = non-negative matrix factorization; PCA = posterior cerebral artery Graphical Abstract cardioembolic stroke subtype.2 In particular, this higher Introduction symptom load emerges despite comparable lesion characteris- Stroke results in a considerable burden of disease for both men tics,3,4 consistent with the possibility that male and female and women. Converging evidence, however, underscores the brains might react differently to ischaemia-induced lesions. relevant influence of biological sex on the idiosyncrasies of More concretely, we recently noted distinct sex divergences cause, presentation and outcome of acute ischaemic stroke in stroke lesion topographies: stroke severity in both men (AIS).1 Of note, women appear to feature a higher AIS severity and women was explained by lesions affecting presumed bi- upon admission that cannot be satisfactorily explained by any lateral motor regions and left-lateralized language regions. sociodemographic or clinical factors, such as age and However, only in women, stroke severity was additionally 4 | BRAIN COMMUNICATIONS 2022: Page 4 of 12 A. K. Bonkhoff et al. explained by lesions extending to the posterior circulation of patterns were incorporated as input variables with hierarchic- the left hemisphere, rendering female-specific lesion pattern al priors capturing the biological sex of participants (as indi- more wide-spread. Whether these sex-specific effects have cated in medical records), that is, one for female and one for an impact on acute stroke symptoms only, or have a long- male participants. The covariates age, age-squared, sex, car- lasting character, is currently unknown. The aim of the pre- diovascular risk factors (history of hypertension, atrial fibrilla- sent study was to address these gaps in our understanding tion, diabetes, ischaemic heart disease, prior stroke and of sex-specific lesion effects on long-term stroke outcomes. smoking) and total DWI-derived lesion volume were included in the model (c.f., Supplementary materials for full model spe- cifications).4,10,11 We did not incorporate admission stroke se- Materials and methods verity as an input variable, given that it likely reflects the extent and location of brain injury. Samples from the AIS patient sample Bayesian posterior parameter distribution were drawn by em- ploying the No U-Turn Sampler (NUTS), a type of Monte All MRI-Genetics Interface Exploration (MRI-GENIE)5 AIS pa- Carlo Markov Chain algorithm (setting: draws = 5000).13 tients with available high-quality diffusion-weighted imaging The model classification performance of unfavourable out- (DWI)-derived lesion segmentations6 and 3 months (60–190 comes was measured by the area under the receiver operating days) modified Rankin Scale (mRS) data were included in this characteristic curve (AUC). We first assessed similarities be- complete case study (c.f., Supplementary materials for a sample tween male- and female-specific lesion pattern effects and de- size calculation). MRI-GENIE is a large international collabor- termined those lesion patterns with posterior distributions ation, built upon the infrastructure of the Stroke Generics substantially differing from zero in both men and women. Network (SiGN).7 It assembled sociodemographic, clinical, Differences between hierarchically estimated female and neuroimaging and genetic data of 3300 AIS patients, with male lesion pattern effects were then evaluated via subtracting the primary aim to facilitate genetic analyses of neuroimaging the corresponding posterior parameter distributions. As in phenotypes as derived from clinical scans.5 While MRI- previous work,4 we presumed substantial lesion pattern ef- GENIE merged data from 12 international sites overall, ana- fects or sex differences if the resulting posterior (difference) lyses here primarily relied on five individual studies that could distributions did not include zero in the 90% highest probabil- additionally share functional outcomes (c.f., Supplementary ity density interval (HPDI). Table 1 for study sizes and characteristics). Subjects gave written informed consent in accordance with the Declaration of Helsinki. The study protocol was approved by the Sensitivity analyses Massachusetts General Hospital’s Institutional Review Board. We evaluated whether there were any general sex differences in lesion anatomy, i.e. the total DWI-derived lesion volume, Stroke lesion representation as well as parcel-wise lesion volumes and frequencies in how derivation often parcels were affected. There was a slight imbalance in the men:women ratio in our DWI data and respective automatically DWI-derived lesions sample. Altogether, there were more male patients experien- were non-linearly spatially normalized (c.f., Wu et al.6 for de- cing a favourable long-term outcome. To ensure that results tails on lesion segmentation and spatial normalization and were not skewed by varying frequencies of unfavourable func- Supplementary materials for imaging parameters). Spatially tional outcomes in women and men, we repeated analyses after normalized DWI data and lesion segmentations were careful- downsampling the majority class, i.e. matching the occurrence ly quality controlled by two experienced raters (A.K.B. and of favourable outcomes in the groups of men and women. M.B.). Lesion data were parsed according to 94 cortical re- More specifically, we downsampled the larger group of men gions, 15 subcortical regions8 and 20 white matter tracts,9 with favourable outcomes (n = 386) to the size of the smaller i.e. the number of lesioned voxels per atlas-defined brain re- group of women with favourable outcomes (n = 208). That gion was computed. Non-negative matrix factorization is, we randomly chose 208 of the 386 men with favourable out- (NMF) was used to reduce the collection of 129 atlas-based comes and repeated this process 20 times. Unfavourable long- region and tract lesion measures to 10 unique lesion patterns. term outcomes were present in the same number of male and The number 10 was chosen in line with previous work4,10–12 female patients (n = 114 each). Afterwards, we repeatedly and further motivated by achieving a balance between faith- conducted the Bayesian logistic regression analyses. ful representation of the high-dimensional input space and mitigation of the risk of overfitting. Lesion network mapping analyses Statistical analyses: modelling Finally, we employed lesion network mapping14 (LNM) aim- ing to uncover links between lesion patterns and sex-specific unfavourable functional outcomes lesion disconnection profiles. To achieve this goal, we imple- Unfavourable functional outcome (mRS . 2) was modelled mented two main changes compared to classic LNM ana- via the Bayesian hierarchical logistic regression. The 10 lesion lyses15,16: first, instead of computing lesion network maps Sex-specific lesion pattern BRAIN COMMUNICATIONS 2022: Page 5 of 12 | 5 for lesions of individual patients, we utilized the 10 prototyp- Atlases. Analyses were implemented in the Python 3.7 (main- ical lesion patterns as regions of interest for the estimation of ly employing the packages nilearn19 and pymc320). whole-brain lesion connectivity. We created 100 variations of Exemplary code is available under: https://github.com/ each of the 10 lesion patterns by sampling a random number AnnaBonkhoff/BMH_functional_outcomes_sex_differences. and collection of lesion pattern-affiliated brain regions (a brain region was considered affiliated, if the NMF-weight ex- ceeded 0.05). Second, we relied on male- and female-specific normative connectomes, i.e. connectomes that were based Results on data of only male or female healthy participants (n = 346 each). Consequently, we obtained two LNM exemplars, a Prediction of unfavourable male- and a female-specific one, for each lesion pattern (con- functional outcomes across men ventionally: one LNM exemplar based on connectomes of n and women = 1000 healthy male and female participants).14 The prepro- cessing of sex-specific resting-state fMRI data itself was per- A total of 822 AIS patients were recruited to model formed as previously described. Global signal regression was unfavourable long-term outcomes [age: 64.7 (15.0) included. After estimating the whole-brain voxel-wise lesion years, 39% women, 28% unfavourable functional outcome, connectivity and Fisher’s z-transformation, we summarized c.f., Table 1 for further clinical characteristics and the positive and negative t-values, i.e. the ‘intensity’ of lesion Supplementary Fig. 1 for a lesion overlap]. Individual atlas- connectivity, within Yeo/Schaefer-defined cortical net- parcellated lesions were represented in five anatomically works17,18 (7 per hemisphere, therefore 14 in total; visual, so- plausible lesion patterns per hemisphere, with varying em- matomotor, dorsal attention, ventral attention, limbic, phases on subcortical to cortical and anterior to posterior re- frontoparietal, default mode network). The Yeo–Schaefer at- gions (Fig. 1). Three of these 10 unique patterns explained las was here chosen to match the functional data at hand, as it considerable variation in unfavourable outcomes concur- was derived from functional data, in contrast to the Harvard– rently in both male and female patients (AUC = 0.81). The Oxford (HO)8 or JHU atlases,9 which are based on structural relevance in explaining unfavourable outcomes across men data. Eventually, we compared network-averaged intensities and women of these three spatial lesion patterns were indi- between the male- and female-connectome-based maps for cated by 90% posterior intervals of Bayesian parameter each lesion pattern (two-sided t-tests, level of significance P distributions excluding zero for both biological sexes , 0.05, false discovery rate-corrected for 14 hemisphere- (men: Lesion pattern #2: posterior mean: 0.20, 90% HPDI: specific network values for 10 lesion patterns). 0.04–0.35; Lesion pattern #7: posterior mean: 0.57, 90% HPDI: 0.11–1.07; Lesion pattern #8: 0.37, 90% HPDI: 0.01–0.72; women: Lesion pattern #2: posterior mean: Code and data availability 0.21, 90% HPDI: 0.07–0.35; Lesion pattern #7: posterior The authors agree to make data available to researchers for mean: 0.76, 90% HPDI: 0.13–1.25; Lesion pattern #8: the explicit purposes of reproducing the here stated results, 0.39, 90% HPDI: 0.07–0.72; Supplementary Figs 2 and 3). pending the permission for data sharing by the Mainly implicated brain regions in these three patterns Massachusetts General Hospital’s institutional review were bilateral subcortical grey matter nuclei and board. The HO and JHU DTI-based white matter atlases left-lateralized regions in proximity to the insula (Figs 1 are openly available at https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/ and 2). Table 1 Patient characteristics All Female participants Male participants participants Statistical comparison of male (n = 822) (n = 500) (n = 332) and female participants Age, years 64.7 (15.0) 63.9 (14.2) 65.8 (16.2) P = 0.07 Sex 39.2% – – — Unfavourable outcome (mRS . 2) 27.7% 22.8% 35.4% P = 0.0001 Normalized DWI-derived stroke lesion volume 3.3 (0–12.8) 2.9 (0–11.3) 3.8 (0–17.8) P = 0.28 (ml, median, interquartile range) Hypertension 64.1% 63.0% 65.9% P = 0.41 Diabetes mellitus Type 2 21.8% 23.0% 19.9% P = 0.30 Atrial fibrillation 16.8% 14.6% 20.2% P = 0.04 Coronary artery disease 18.4% 21.8% 13.0% P = 0.002 Smoking 55.0% 61.0% 45.7% P , 0.0001 Prior stroke 9.7% 9.4% 10.2% P = 0.72 Mean values and standard deviation, unless otherwise noted. Characteristics of men and women were compared either via two-sample t-tests or two-sided Fisher’s exact tests as appropriate. Significantly more women than men experienced unfavourable outcomes obtained 3 months post-stroke (women: 35.4% versus men: 22.8%). In view of this difference in our main outcome, we performed additional downsampling analyses, in which we repeatedly contrasted samples of male and female patients with the same ratios of favourable to unfavourable outcomes. 6 | BRAIN COMMUNICATIONS 2022: Page 6 of 12 A. K. Bonkhoff et al. Figure 1 Anatomically plausible, parsimonious representation of stroke lesions. Ten unique, archetypical stroke lesion patterns were derived via an unsupervised pattern-discovery framework. Lesion pattern represented predominantly right-hemispheric stroke (A) and left-hemispheric stroke (B) with varying emphases on cortical–subcortical and anterior–medial–posterior regions. Three lesion patterns, framed in yellow for right-hemispheric stroke and in orange for left-hemispheric stroke, had a high relevance in explaining unfavourable functional outcome 3 months after stroke for both men and women. This relevance was discernible from their Bayesian posterior distributions that did not substantially overlap with zero (i.e. their 90% credibility intervals did not include zero). Sex-specific prediction of #10: −0.30, 90% HPDI: −0.56 to −0.07, Fig. 3, c.f., Supplementary Figs 2 and 3 for an exhaustive display of unfavourable functional outcomes difference distributions]. Next, we compared the relevance of lesion patterns in men versus in women via contrasting their respective sex- Sensitivity analyses indicate stability specific lesion pattern posterior distributions. We here ob- served one predominant sex difference: the left-hemispheric of results lesion pattern of posterior circulation brain regions was as- Total lesion volume, parcel-wise lesion volumes, as well as signed a substantially higher relevance in explaining un- parcel-wise frequencies of lesion status did not significantly favourable functional outcome specifically in women differ between men and women (P . 0.05, family wise error- [mean difference (men – women) of Bayesian marginal pos- corrected, Supplementary Table 2). Results remained un- terior distributions related to expressions of Lesion pattern changed after reiterating the identical analysis workflow in Sex-specific lesion pattern BRAIN COMMUNICATIONS 2022: Page 7 of 12 | 7 Figure 2 Individual brain regions explaining unfavourable functional outcomes. Characteristic constellations of cortical (A) and subcortical brain regions, as well as white matter tracts (B) emerged that explained unfavourable outcomes 3 months after stroke in 500 male and 322 female stroke patients. Lesions in the left hemisphere were more strongly associated with unfavourable long-term outcomes than lesions in the right hemisphere for both men and women. Particularly relevant regions comprised left pre- and post-central, insular and opercular cortex, superior and middle temporal gyri, supramarginal and angular gyrus and lateral occipital cortex. balanced groups of men and women with favourable out- thus reinforced the notion that detected sex differences comes: In the 20 logistic regression analyses resulting after were not due to imbalances in the men:women ratio or downsampling, we consistently observed female-specific more favourable long-term outcomes in men. Lesion pattern #10 effects, i.e. the difference distribution of Lesion pattern #10 posteriors did not overlap with zero in 16 cases (80% of cases). In the remaining four downsampling Pronounced lesion connectivity scenarios, the difference distributions had a minor overlap disturbances in women in Lesion with zero. Importantly, only the female-specific Lesion pat- tern #10 posterior distributions indicated a substantial effect, pattern #10 while the male-specific Lesion pattern #10 distributions over- Lesion pattern #10 was characterized by the most pro- lapped with zero. Altogether, these downsampling analyses nounced differences in lesion connectivity as uncovered by 8 | BRAIN COMMUNICATIONS 2022: Page 8 of 12 A. K. Bonkhoff et al. Figure 3 Sex-specific effects relating to left-hemispheric posterior circulation lesion pattern. (A) Lesion pattern #10 represented left-hemispheric lesions in the presumed posterior circulation, combining the hippocampus, precuneus, fusiform and lingual gyrus, latero-occipital cortex and occipital pole. (B) The difference distribution of Bayesian posteriors for the male- and female-specific expression of Lesion pattern #10 indicated a substantially higher relevance in explaining unfavourable functional outcome specifically in women (mean of the difference posterior distribution = −0.295, HPDI of the posterior distribution covering 90% certainty = −0.556 to −0.068). LNM (Fig. 4). More specifically, Lesion pattern #10 featured aggressive acute reperfusion treatments, special post-acute re- nine significant male- and female-specific intensity differ- habilitative efforts and a sex-specific planning of care. ences, that were consistent with women expressing more in- Sex divergences in acute symptoms and long-term out- tense lesion connectivity. Only one other lesion pattern comes post-stroke are frequently examined with respect to (Lesion pattern #2) had the same number of significant sociodemographic and clinical aspects: For example, previ- sex-specific intensity differences. Here, higher intensity how- ous studies have suggested that women experience a higher ever occurred for male- and female-based connectivity. stroke severity partly due to their more advanced age and Altogether, these explorations thus suggest that sex differ- higher likelihood of pre-morbid disability, as well as higher ences, as ascertained for Lesion pattern #10, might be linked frequency of atrial fibrillation and cardioembolic strokes.2,21 to differences in affected functional connectivity. In women, However, the importance of actual lesion characteristics has Lesion pattern #10 was associated with a more pronounced rarely been studied in a sex-specific manner, that is lesion ef- deterioration in functional connectivity than in men. fects in men and women have rarely been compared, neither in low dimensions (e.g. total lesion volume) nor in high- dimensional settings (e.g. spatially high-resolution lesion- symptom mapping studies).22 In most cases, if considered Discussion at all, the biological sex of patients is regressed out from le- Our study provides evidence that left-lateralized posterior cir- sion information,23 essentially rendering it impossible to culation strokes increase the likelihood of unfavourable subsequently deduce any linear sex-specific effects. In con- 3-month outcomes specifically in women. This effect was trast, the here installed Bayesian hierarchical modelling not found in male AIS patients. Our present findings thereby framework allowed for the estimation and systematic com- support and extend previous inferences on sex-specific lesion parison of male- and female-specific lesion effects. Our find- topographies for acute stroke severity4: lesions affecting the ings enabled by these sex-specific pattern analyses now (i) left hippocampus, precuneus, lingual and fusiform gyrus, offer the potential to enhance stroke outcome predictions the occipital pole and latero-occipital cortex, do not only ex- and (ii) motivate the focused investigation of sex-stratified plain more severe strokes in women in the acute phase. stroke care decisions with the ultimate goal of improving Rather, they have an additional permanent female-specific ef- long-term outcomes. fect and explain unfavourable outcomes in a later phase after Neuroanatomically, it is particularly striking that ob- stroke. Future studies are warranted to now test whether wo- served sex differences pertained to the left hemisphere men with these brain regions at risk may benefit from more only. Rare examples of early sex-specific lesion studies Sex-specific lesion pattern BRAIN COMMUNICATIONS 2022: Page 9 of 12 | 9 Figure 4 Significantly altered male- and female-connectome-based lesion connectivity in 14 cortical networks (seven per hemisphere). Sex-specific lesion connectivity was computed for each of the 10 unique lesion patterns (c.f., Fig. 1) and subsequently statistically compared within each of the cortical networks. Networks with significantly different lesion connectivity are represented in colour (orange/red: significantly stronger in women; blue: significantly stronger in men). To allow for these statistical group comparisons in the first place, we inserted an additional simulation step: each lesion pattern was slightly varied 100 times by sampling a random number and collection of lesion pattern-affiliated brain regions. Exemplarily more in detail: Lesion pattern #1 primarily comprised the parcels pre- and post-central gyrus, superior, middle or inferior frontal gyrus, insular cortex, superior parietal and supramarginal cortex. We would here, for example, randomly choose pre- and post-central gyrus for a first Lesion pattern #1-like lesion, subsequently choose superior parietal, as well as supramarginal cortex for a second Lesion pattern #1-like lesion and so forth until we obtained 100 Lesion pattern #1-like lesions. Male- and female-specific lesion connectivity was then computed for each of these 100 simulated lesions per lesion pattern. Lesion patterns #2 and #10 comprised most connectivity differences (nine each). While Lesion pattern #2 was characterized by both higher connectivity in men and women, Lesion pattern #10 featured higher lesion connectivity exclusively in women. already indicated that left-sided lesions may have compara- is thought to be involved in the identification of words tively more detrimental effects in women.24,25 These studies from their visual shape and thus to contribute to high-level contrasted left- and right-hemispheric strokes and their ef- processing of word meaning.27 Collectively, these previous fects on intelligence measures in men and women. They con- insights may motivate the hypothesis that the left occipital cluded that women experience deterioration of both verbal fusiform gyrus, as encompassed in our female-specific left and performance scale IQ values after left-hemispheric le- posterior circulation lesion pattern, could be one of the driv- sions, while in men and in the case of right-hemispheric ing forces for our observed sex divergences. Nonetheless, a lesions just one of these qualities was affected.24,25 direct comparison of these earlier findings and ours is ham- Additionally, a recent functional connectivity-based study pered by the crudeness of our spatial lesion representation on sex-stratified predictions of IQ scores reported a high pre- and choice of outcome score. dictive capacity of fusiform gyrus-related functional con- Therefore, the coarse-grained nature of our 10 lesion nectivity patterns specifically in women.26 Of note, the left patterns, that each combined several individual brain re- fusiform gyrus comprises the ‘visual word form area’ that gions, as well as of our global 3-month clinical endpoint, 10 | BRAIN COMMUNICATIONS 2022: Page 10 of 12 A. K. Bonkhoff et al. unfavourable outcome as measured on the mRS, can be more sex-specific approach to the planning of acute stroke considered limitations of this study. Nonetheless, the spa- trials in the shorter and clinical stroke management in the tial granularity of our lesion patterns goes beyond one of longer term. Our findings suggest that women could, exem- the early intelligence-focused lesion studies that differen- plarily, benefit from more aggressive acute treatments of dis- tiated between left- and right-hemispheric strokes tal PCA occlusions. These more sex-informed clinical only.24,25 Anatomical configurations of stroke lesions are practices may eventually augment AIS outcomes on a broad furthermore naturally constrained by the human vascula- population basis. ture and our data-driven dimensionality reduction ap- proach addresses this circumstance overtly by combining brain regions that tend to be lesioned simultaneously. Furthermore, the here-derived low-dimensional lesion re- Acknowledgements presentations have been shown to be highly reproducible We are grateful to our colleagues at the J. Philip Kistler in independent datasets in previous work.4 The mRS cap- Stroke Research Center for valuable support and discus- tures a patient’s functional status on a very general level sions. Furthermore, we are grateful to our research partici- only, but is frequently used as a primary outcome measure pants without whom this work would not have been in large clinical stroke trials.28 It may thus represent a nat- possible. ural first choice; especially, as ascertaining sex differences in lesion topographies for such a global outcome automat- ically suggests a meaningful impact on a broad stroke population basis. Future studies are, however, needed to pinpoint the exact brain correlates and brain functions Funding of the detected female-specific effects and to furthermore A.K.B. is supported by the Massachusetts General Hospital disentangle the effects of pre-stroke functional states and Executive Committee on Research (MGH ECOR) Fund for acute recanalization therapies. For example, a concrete Medical Discovery (FMD) Clinical Research Fellowship first step could be to conduct prospective stroke trials to Award. M.B. acknowledges support from the Société systematically evaluate sex-specific treatment decisions, Française de Neuroradiologie, Société Française de such as more aggressive endovascular reperfusion therap- Radiologie and Fondation ISITE-ULNE. A.V.D. is in part ies of the posterior cerebral artery (PCA) occlusions in wo- supported by the National Institutes of Health and the men. Especially since thrombectomy for primary distal National Institute of Neurological Disorders and Stroke PCA occlusion stroke was recently determined to be safe (NIH-NINDS, R01 NS103824, RF1 NS117643, R01 and potentially beneficial,29 we would hypothesize a re- NS100417, U01NS100699 and U01NS110772). C.J.G. ac- sulting improvement in outcomes in view of our present knowledges support from the Swedish Research Council findings. Importantly, posterior circulation strokes occur (2018-02543), the Swedish state under the agreement be- in 30% of all cases across both men and women30 and tween the Swedish government and the county councils, approximately 13% of all strokes more specifically relate the ‘Avtal om Läkarutbildning och Medicinsk Forskning’ to PCA occlusions,31 underscoring the potential benefit (ALF) agreement (ALFGBG-720081); the Swedish Heart of enhanced sex-specific treatment outcomes. Finally, our and Lung Foundation (20190203). A.G.L. acknowledges performance estimates were computed based on the support from the Swedish Research Council (2019-01757), Bayesian posterior predictive distributions, as analyses The Swedish Government (under the ‘Avtal om were performed in a Bayesian hierarchical regression Läkarutbildning och Medicinsk Forskning, ALF’), The framework. Especially if future work shifted the focus Swedish Heart and Lung Foundation, Region Skåne, Lund from inference on sex differences, as applied here, to a fo- University, Skåne University Hospital, Sparbanksstiftelsen cus on clinical outcome prediction,32,33 this future work Färs och Frosta, Fremasons Lodge of Instruction Eos in could complement our work by running cross-validated Lund and the National Institutes of Health (NIH, out-of-sample predictions. We here aimed to mitigate the 1R01NS114045-01). D.B. has been funded by the Brain risk of overfitting by investigating a limited number of in- Canada Foundation, through the Canada Brain Research put variables in a large sample of 800 stroke patients. Fund, with the financial support of the Health Canada, the The Bayesian priors may moreover intrinsically exert National Institutes of Health (NIH R01 AG068563A), some regularizing effect. the Canadian Institute of Health Research (CIHR 438531), Our study brings into sharp focus the possibility that the Healthy Brains Healthy Lives initiative (Canada First lesions in the left posterior circulation underlie a female- Research Excellence fund), the Google (Research Award, specific higher likelihood of unfavourable functional Teaching Award) and by the Canada Institute for outcomes approximately 3 months after AIS. Of note, we Advanced Research (CIFAR) Artificial Intelligence Chairs did not observe any sex differences in lesion volumes per se program. N.S.R. is in part supported by the National and findings remained the same when controlling for varying Institutes of Health and the National Institute of rates in unfavourable outcomes between men and women. Neurological Disorders and Stroke (NIH-NINDS, If confirmed in future research, these findings prompt a R01NS082285, R01NS086905 and U19NS115388). Sex-specific lesion pattern BRAIN COMMUNICATIONS 2022: Page 11 of 12 | 11 intracranial large-artery occlusion. Cerebrovasc Dis. 2010;30(5): Competing interests 470–475. 4. Bonkhoff AK, Schirmer MD, Bretzner M, et al. Outcome after acute M.R.E. has received personal fees for consulting from Astra ischemic stroke is linked to sex-specific lesion patterns. Nat Zeneca and WorldCare Clinical Group. C.J.G. has received Commun. 2021;12(1):3289. consulting honoraria from Microvention and Strykere, and 5. Giese AK, Schirmer MD, Donahue KL, et al. Design and rationale research funding from Medtronic and Penumbra. A.V. has for examining neuroimaging genetics in ischemic stroke: The received research funding from Cerenovus. A.G.L. has re- MRI-GENIE study. Neurol Genet. 2017;3(5):e180. 6. Wu O, Winzeck S, Giese AK, et al. Big data approaches to phenotyp- ceived personal fees from Bayer, Astra Zeneca, BMS Pfizer ing acute ischemic stroke using automated lesion segmentation of and Portola. N.S.R. has received compensation as scientific multi-center magnetic resonance imaging data. Stroke. 2019;50: advisory consultant from Omniox, Sanofi Genzyme and 1734–1741. AbbVie Inc. All other authors declare no competing 7. Meschia JF, Arnett DK, Ay H, et al. Stroke Genetics Network interests. (SiGN) study: Design and rationale for a genome-wide association study of ischemic stroke subtypes. Stroke. 2013;44(10):2694–2702. 8. Desikan RS, Ségonne F, Fischl B, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans Supplementary material into gyral based regions of interest. Neuroimage. 2006;31(3): 968–980. Supplementary material is available at Brain Communications 9. Mori S, Wakana S, Van Zijl PC, Nagae-Poetscher LM. MRI atlas of human white matter. Elsevier; 2005. online. 10. Bonkhoff AK, Lim J-S, Bae H-J, et al. Generative lesion pattern de- composition of cognitive impairment after stroke. Brain Commun. 2021;3(2):fcab110. Appendix I 11. Bonkhoff AK, Hong S, Bretzner M, et al. Lesions in putative language and attention regions are linked to more severe strokes in patients with higher white matter hyperintensity burden. medRxiv. 2021. MRI-GENIE and GISCOME 12. Regenhardt RW, Bonkhoff AK, Bretzner M, et al. Association of in- farct topography and outcome after endovascular thrombectomy in Investigators and the International patients with acute ischemic stroke. Neurology 2022. Stroke Genetics Consortium 13. Hoffman MD, Gelman A. The No-U-Turn sampler: Adaptively set- ting path lengths in Hamiltonian Monte Carlo. J Mach Learn Res. Anna K. Bonkhoff, Martin Bretzner, Sungmin Hong, 2014;15(1):1593–1623. Markus D. Schirmer, Alexander Cohen, Robert 14. Fox MD. Mapping symptoms to brain networks with the human W. Regenhardt, Kathleen L. Donahue, Marco J. Nardin, connectome. N Engl J Med. 2018;379(23):2237–2245. Adrian V. Dalca, Anne-Katrin Giese, Mark R. Etherton, 15. Ferguson MA, Lim C, Cooke D, et al. A human memory circuit de- rived from brain lesions causing amnesia. Nat Commun. 2019; Brandon L. Hancock, Steven J. T. Mocking, Elissa 10(1):1–9. C. McIntosh, John Attia, Oscar R. Benavente, Stephen 16. Cohen AL, Soussand L, Corrow SL, Martinaud O, Barton JJ, Fox Bevan, John W. Cole, Amanda Donatti, Christoph MD. Looking beyond the face area: Lesion network mapping of J. Griessenauer, Laura Heitsch, Lukas Holmegaard, prosopagnosia. Brain. 2019;142(12):3975–3990. Katarina Jood, Jordi Jimenez-Conde, Steven J. Kittner, 17. Yeo BT, Krienen FM, Sepulcre J, et al. The organization of the hu- man cerebral cortex estimated by intrinsic functional connectivity. J Robin Lemmens, Christopher R. Levi, Caitrin Neurophysiol. 2011;106:1125–1165. W. McDonough, James F. Meschia, Chia-Ling Phuah, 18. Schaefer A, Kong R, Gordon EM, et al. Local-global parcellation of Arndt Rolfs, Stefan Ropele, Jonathan Rosand, Jaume the human cerebral cortex from intrinsic functional connectivity Roquer, Tatjana Rundek, Ralph L. Sacco, Reinhold MRI. Cereb Cortex. 2018;28(9):3095–3114. Schmidt, Pankaj Sharma, Agnieszka Slowik, Martin 19. Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: Machine learning in Python. J Mach Learn Res. 2011;12:2825–2830. Söderholm, Alessandro Sousa, Tara M. Stanne, Daniel 20. Salvatier J, Wiecki TV, Fonnesbeck C. Probabilistic programming in Strbian, Turgut Tatlisumak, Vincent Thijs, Achala Vagal, Python using PyMC3. PeerJ Comput Sci. 2016;2:e55. Johan Wasselius, Daniel Woo, Ramin Zand, Patrick 21. Dehlendorff C, Andersen KK, Olsen TS. Sex disparities in stroke: F. McArdle, Bradford B. Worrall, Christina Jern, Arne Women have more severe strokes but better survival than men. J G. Lindgren, Jane Maguire, Michael D. Fox, Danilo Am Heart Assoc. 2015;4(7):e001967. 22. Bates E, Wilson SM, Saygin AP, et al. Voxel-based lesion–symptom Bzdok, Ona Wu and Natalia S. Rost. mapping. Nat Neurosci. 2003;6(5):448–450. 23. Wu O, Cloonan L, Mocking SJT, et al. Role of acute lesion topog- raphy in initial ischemic stroke severity and long-term functional References 24. outcomes. Stroke. 2015;46(9):2438–2444. Inglis J, Ruckman M, Lawson JS, MacLean AW, Monga TN. Sex 1. Bushnell C, Howard VJ, Lisabeth L, et al. Sex differences in the differences in the cognitive effects of unilateral brain damage. evaluation and treatment of acute ischaemic stroke. Lancet Cortex. 1982;18(2):257–275. Neurol. 2018;17(7):641–650. 25. Inglis J, Ruckman M, Lawson JS, MacLean AW, Monga TN. Sex 2. Bonkhoff AK, Karch A, Weber R, Wellmann J, Berger K. Female differences in the cognitive effects of unilateral brain damage: stroke: Sex differences in acute treatment and early outcomes of Comparison of stroke patients and normal control subjects. acute ischemic stroke. Stroke. 2021;52(2):406–415. Cortex. 1983;19(4):551–555. 3. Silva GS, Lima FO, Camargo ECS, et al. Gender differences in out- 26. Jiang R, Calhoun VD, Fan L, et al. Gender differences in comes after ischemic stroke: Role of ischemic lesion volume and connectome-based predictions of individualized intelligence 12 | BRAIN COMMUNICATIONS 2022: Page 12 of 12 A. K. Bonkhoff et al. quotient and sub-domain scores. Cereb Cortex. 2020;30(3): 30. Frid P, Drake M, Giese AK, et al. Detailed phenotyping of posterior 888–900. vs. anterior circulation ischemic stroke: A multi-center MRI study. 27. Dehaene S, Cohen L. The unique role of the visual word form area in J Neurol. 2020;267(3):649–658. reading. Trends Cogn Sci. 2011;15(6):254–262. 31. Drake M, Frid P, Hansen BM, et al. Diffusion-weighted imaging, MR 28. Thomalla G, Simonsen CZ, Boutitie F, et al. MRI-guided thromb- angiography, and baseline data in a systematic multicenter analysis of olysis for stroke with unknown time of onset. N Engl J Med. 3,301 MRI scans of ischemic stroke patients—neuroradiological re- 2018;379(7):611–622. view within the MRI-GENIE study. Front Neurol. 2020;11:577. 29. Meyer L, Stracke CP, Jungi N, et al. Thrombectomy for primary dis- 32. Shmueli G. To explain or to predict? Stat Sci. 2010;25(3):289–310. tal posterior cerebral artery occlusion stroke: The TOPMOST 33. Bzdok D, Ioannidis JPA. Exploration, inference, and prediction in study. JAMA Neurol. 2021;78:434–444. neuroscience and biomedicine. Trends Neurosci. 2019;42(4):251–262.