11Anxiety outcomes following traumatic brain injury (TBI) are complex, and the underlying neural 12 mechanisms are poorly understood. Here, we developed a multidimensional behavioral profiling 13 approach to investigate anxiety-like outcomes in mice that takes into account individual variability. 14 Departing from the tradition of comparing outcomes in TBI versus sham groups, we identified animals 15 within the TBI group that are vulnerable to anxiety dysfunction by applying dimensionality reduction, 16 clustering and post-hoc validation to behavioral data obtained from multiple assays for anxiety at 17 several post-injury timepoints. These vulnerable animals expressed distinct molecular profiles in the 18 corticolimbic network, with downregulation in GABA and glutamate, and upregulation in NPY 19 markers. Indeed, among vulnerable animals, not resilient or sham controls, severity of anxiety 20 outcomes correlated strongly with expression of molecular markers. Our results establish a 21 foundational approach, with predictive power, for reliably identifying maladaptive anxiety outcomes 22 following TBI and uncovering neural signatures of vulnerability to anxiety.23 24 101 sections (Methods).102 103 Figure 1. Multidimensional behavioral clustering with validation (MBCV) for identifying animals 104 vulnerable to anxiety after TBI 105 (A) Experimental timeline and protocol for measuring anxiety-like behaviors before and after injury. EZM 106 -elevated zero maze; OFT -open field test; EPM -elevated zero maze; all standard assays for 107 5 measuring anxiety-like behaviors in rodents. CCI -Controlled cortical impact injury model. IHC -108 immunohistochemistry. 109 (B) Anxiety behavioral 'profile' of a mouse: A 11-dimensional vector which combines behavioral data on 110 the EZM, OFT and EPM across time points. Each element in the vector is the proportion of time spent by 111 the mouse in the anxiogenic zones of one of the assays (ex. EZM) at one of the post-injury time points 112 (ex. Week 7), normalized to the animal's pre-injury baseline value in that assay (Methods).
113(C) Multidimensional behavioral clustering with validation approach: Behavioral profiles of mice exposed 114 TBI are first subject to principal components analysis (PCA), and then to k-means clustering (with k=2) to 115 identify two clusters in principal component (PC) space. The final validation step determines (i) whether 116 animals in the two clusters exhibit distinct anxiety phenotypes after TBI, and if so, (ii) whether (and 117 which) cluster consists of animals vulnerable to anxiety outcomes after TBI. This involves plotting and 118 comparing between the clusters, anxiety metrics from each assay.
120We asked if, based on their multidimensional anxiety behavioral profiles, mice that underwent TBI could 121 be separated into two distinct groups such that one exhibited severe affective behavioral consequences 122 of injury, and the other was largely resistant to effects of injury. To address this question, we developed 123 a data-driven approach. W...