Ultrafast ultrasound imaging pattern analysis reveals distinctive dynamic brain states and potent sub-network alterations in arthritic animals

Figures

• 

  Figure 1. Experimental design. (A) Time line of the experimental design. Behaviour habituation and tests were performed during the development of the model. A thinned-skull window was made four days before the imaging session. (B) Setup of the fUS imaging experimentation. The rat was placed on a stereotaxic frame over a warm plate and is continuously perfused with the anesthetic mixture ketamine/medetomidine via a syringe pump. Its arterial pressure, as well as heart and respiratory rates were measured periodically. The ultrasound probe was positioned above the previously thinned skull, with echographic gel in between. The probe was moved using the 4-axis motor. (C) 11 tilted planar ultrasonic waves were emitted into the rat brain. The coherent sum of the resulting 11 images produced a compound image every 2 ms. The final Doppler image was the mean image over a 400 ms acquisition period, i.e., 200 compound images. The CBV signal was then extracted using a “clutter” filter that removed the tissue signal, leaving the blood signal intact. Computing the signal of each pixel in the plane over ten minutes, we correlated them two-by-two to study FC (D) Vascular signal of the planes imaged during…

• 

  Figure 2. Arthritic rats display (3 weeks after the induction of the model) a sustained peripheral inflammation, reduced body weight gain and mechanical hypersensitivity. Clinical (A,B) and behavioural (C,D) alterations in arthritic animals. (A) Weight gain (g) between week 3 and week 0. (B) Global inflammation score showing a strong inflammation in arthritic animals (from 0 to 60). (C) Reduction of the threshold of static mechanical hypersensitivity, suggesting a bilateral mechanical hypersensitivity in arthritic rats. (D) Measure of the vocalizations evoked by bending of the hind paw (bending score from 0 to 5) shows a strong bilateral hypersensitivity to paw mobilization. N = 9 per group. ***p < 0.001 (U-Mann Whitney test).

• 

  Figure 3. Seed-based analysis of the FC alterations in arthritic animals. (A–E) are typical examples of the correlation between a seed region (ROI delineated in cyan) and the pixels in the imaged plane. The Pearson correlation was then computed in the ROI (delineated in black). The results in (A–E) and their quantification in N = 9 animals per group show that arthritic animals display a reduced bilateral connectivity in the somatomotor (p = 0.002, A), cingulo-cortical (p = 0.02, B), cortico-hippocampal (p = 0.01, C) and cortico-orbital (p = 0.02, D) networks. (F) Boxplots of the correlation coefficients for each group of rats (N = 9 per group) and each pair of ROIs (SM = Somatomotor cortex, Cg = Cingulate cortex, Hipp = Hippocampus, LO = Lateral orbital cortex). *p < 0.05 and **p < 0.01, Welch’s test, or Mann-Whitney test in case of non-normality after Fisher transformation, followed by Benjamini-Hochberg’s correction for multiple comparisons.

• 

  Figure 4. Correlation matrix analysis of the FC on the −0.6 Bregma plan of the rat brain atlas. (a) Plane of interest of the Paxinos rat brain atlas (Bregma −0.6 mm) and corresponding ROIs (S1HL, M1, M2, Cg1, Cg2 left and right for each), with permission from Elsevier. This image is modified from ‘Rat Brain in Stereotaxic Coordinates’. Paxinos &Watson, Academic Press, San Diego 3rd, (1997). (b) Doppler image of the rat brain with the corresponding mask of the rat brain atlas presented in (a). (c,d) Averaged Pearson correlation matrix for each of the 9 rats in the control and arthritic groups. (e) Matrix of significance of the differences between c (control) and d (arthritic) matrices. White squares represent the pairs of ROIs with a significant alteration in FC between the two groups, with corresponding p-values. (f) Boxplot representation of each pair with a significant FC alteration. N = 9 per group. *p < 0.05, **p < 0.01 and ***p < 0.001, Welch’s test, or Mann-Whitney test in case of non-normality after Fisher transformation, followed by Benjamini-Hochberg’s correction for multiple comparisons.

• 

  Figure 5. Dynamic brain state changes in control and arthritic rats imaged in the plan Bregma −0.6 mm. (A,B) Time course of the L1 norm of the phase-locked matrix of a control and an arthritic rat, respectively. Inserted matrices show brain phase patterns at different time points. As previously described by Barttfeld et al.9 the time series show an exploration of brain states over time. (C) Decomposition into brain states 5, 6 and 7 obtained by unsupervised k-means clustering of the phase matrices. The ROIs are the same as in Fig. 4. The decomposition produces robust results, with the matrices obtained for k = 5, 6 or 7 states being very consistent. The p-values written under each matrix show that for four states (states 1, 2, 3 and 7), the occurrence over time is significantly altered between controls and arthritics. (D) Box plots presenting the occurrence rate of each of the four different states depending on the decomposition value k. The first modified state appears significantly more often over time in the control group than in the arthritic group. The other three states appear less frequently over time in the control group than in the arthritic group. While arthritic animals spend…

• 

  Figure 6. Statistically significant correlations between individual physiological or behaviour alterations and altered FC (A,B) or brain states (C). (A,B) Spearman correlation of the 4 and 32 significant results from the seed-based (Bregma −0.6 mm and the two lateral planes, as shown on Fig. 3) (A) or correlation matrix (Bregma −0.6 mm, as shown on Fig. 4) analyses (B), respectively, with behaviour. (C) Correlation of behaviour with the occurrence probability of each state from the k-means decomposition into 7 states. For each type of analysis (seed-based, correlation matrix and FC dynamics), we correlated all individual results (from all control and arthritic rats) with the individual behavioural data. Following a Spearman correlation computation, the results show only the statistically significant correlations (p < 0.05). The associations between FC/brain states and FC/brain states are outlined with a blue square; correlations between FC/brain states and behaviour is indicated by a yellow square; and finally, associations of behaviour with behaviour are outlined with a purple square. This analysis shows specific positive or negative correlations between either brain states or changes in…

• 

  Figure 7. ROC curves derived from the FC data using the seed-based analysis of four statistically significant pairs of ROIs imaged at different planes in control and arthritic rats. The figure shows the increase in sensitivity (or true positive rate) and specificity (or true negative rate) when discriminating between the ‘control’ and ‘arthritic’ conditions obtained when the number of ROI pairs is increased. The individual pairs of ROIs computed were (i) Cg L and SM cortex R (imaged at Bregma −0.6 mm); (ii) Cg L and SM cortex R + SM cortex L/R (imaged at Bregma −0.6 mm); (iii) Cg L and SM cortex R + SM cortex L/R (imaged at Bregma −0.6 mm) + hippocampus and neocortex (imaged at Lateral +0.4 mm); (iv) Cg L and SM cortex R + SM cortex L/R (imaged at Bregma −0.6 mm) + hippocampus and neocortex (imaged at Lateral +0.4 mm) + LO and SM cortices (imaged at Lateral +2.9 mm). The use of four FC pairs leads to an AUC of 0.975 and an optimal discrimination between arthritic and control rats with 100% sensitivity and 88.9% specificity (A,B). (C) Plotting of the sensitivity and specificity obtained using these four FC pairs and the correlation threshold obtained. (D) Cross validation of this approach…