Reliability of fNIRS for noninvasive monitoring of brain function and emotion in sheep

The aim of this work was to critically assess if functional near infrared spectroscopy (fNIRS) can be profitably used as a tool for noninvasive recording of brain functions and emotions in sheep. We considered an experimental design including advances in instrumentation (customized wireless multi-distance fNIRS system), more accurate physical modelling (two-layer model for photon diffusion and 3D Monte Carlo simulations), support from neuroanatomical tools (positioning of the fNIRS probe by MRI and DTI data of the very same animals), and rigorous protocols (motor task, startling test) for testing the behavioral response of freely moving sheep. Almost no hemodynamic response was found in the extra-cerebral region in both the motor task and the startling test. In the motor task, as expected we found a canonical hemodynamic response in the cerebral region when sheep were walking. In the startling test, the measured hemodynamic response in the cerebral region was mainly from movement. Overall, these results indicate that with the current setup and probe positioning we are primarily measuring the motor area of the sheep brain, and not probing the too deeply located cortical areas related to processing of emotions.


Supplementary Section S1 Neuro-anatomical imaging of the sheep's head
We guided the positioning of the fNIRS probe by MRI data of the very same animals, and checked the functional origin of the signal by diffusion tensor imaging (DTI) MRI on the animal brain.
Preliminary MR scans were used to obtain relative position of the brain within the surrounding structures of the head. The MR data were acquired by means of an ESAOTE VETSCAN MR-Grande operating at 0.25 T. T1 weighted images were obtained with 2D Spin Echo T1 sequences with the following parameters: Slice Thickness = 3 mm, Repetition Time TR = 980 ms, EchoTime TE = 18 ms, Number of Averages: 2, Field of View FOV = 230x230 mm 2 . A total of 27 slices were acquired. T2 weighted images were acquired through a 2D Fast Spin Echo T2 sequence with the following parameters: Slice Thickness: 3 mm, Repetition Time TR = 6030 mm, Echo Time TE = 100 ms, Number of Averages: 2, Field of View = 230x230 mm 2 . A total of 27 slices were acquired.
For DTI we utilized the brains of six adult sheep, whose heads were collected at a commercial abattoir during routine slaughtering procedures. Slaughtering was performed according to the European Community Council directive (86/609/EEC) that regulates animal welfare during the whole commercial process and guarantees that animals are treated humanely and constantly monitored under mandatory official veterinary medical care. Once removed from the head, the brains were immediately fixed by immersion in cold buffered formalin. The time interval between death and removal of the brain varied between 10 and 20 minutes. The fixed brains were subsequently transported to the Department of Computer Science of the University of Verona for MR scans, using a 4.7 Tesla (T) magnet. Images were acquired with an Echo Planar Imaging (EPI) sequence with the following parameters: TR 20000 ms, TE 24.7 ms, FOV = 6.0x5.0 cm 2 ; MTX 120x100; isotropic in-plane resolution of 0.500 mm; slice thickness 1.0 mm; number of slice 80; EPI factor 11; NEX 6; 30 non-collinear directions acquired with a b-value of 3000 s/mm 2 and 5 b0 images for a total acquisition time of about 12 h 50 min (for further details and discussion of motor projections see Ref. 42 ).
From MRI and DTI data we confirmed that the location of the fNIRS probe was over the motor area of the cortex (see Supplementary Figure SF1).  at 839 nm, as reported in in Supplementary Table ST1. We have also derived the values for the DPF at 739 nm and 851 nm by fitting the data at 671, 730, 780, and 830 nm to a quadratic polynomial function obtaining 6.6 and 6.0 at 751 nm and 839 nm, respectively.

Supplementary Section S3 Depth sensitivity
In fNIRS studies, task related and task unrelated physiological changes occurring in the scalp can introduce confounding signals often leading to false positives and artefacts 54 . The adoption of a multi-distance approach is therefore essential in CW fNIRS to distinguish signals from shallow (e.g. scalp, skull and CSF) and deep (e.g. gray matter and white matter) layers in the head. This approach is justified by the physics of photon migration in diffusive media: in steady state (i.e. CW) the longer the source detector distance, the deeper is the average penetration depth 55 . Given an average value of scalp to cortex distance of about 15 mm 56 in fNIRS studies on adult human subjects, channels with relatively short source detector distance (e.g. ρ ≤ 10 mm) are typically only able to reach surface layers, while channels with relatively long separation (e.g. ρ ≥ 30 mm) can typically reach deeper layers in the head such as gray matter 57 .
It has been reported that the cerebral cortex of sheep is located about 5 to 9 mm below the scalp 25 . According to anatomical and MRI measurements on the heads of several sheep of the same age (see Supplementary Section S1), for this study we could assume that the scalp and skull thickness of the sheep is on average 10 mm under the fNIRS probe. Therefore, in this study we used ρ = 10 mm and ρ = 30 mm as short and long channels, respectively. A shorter short distance (e.g. ρ = 5 mm) could not be used due to constraints in the shape of light emitter and detector of the CW fNIRS device we used. The use of a distance longer than 30 mm was prevented by the available space on the sheep head.
In order to have a rough estimate of the penetration depth of CW NIRS in our study, we For a 30 mm thick homogenous slab with μ a = 0.178 cm -1 and μ s ′ = 11.9 cm -1 , the mean maximum depths are <Zmax| ρ = 10 mm> ≈ 5.48 mm and <Zmax| ρ = 30 mm> ≈ 11.60 mm.
To provide a representation of photon path in a diffusive two-layer medium we calculated the sensitivity maps for ρ = 10 mm and ρ = 30 mm by means of a perturbation model 58 .
Assuming that the thickness of the upper (extra-cerebral) layer is 10 mm, we can clearly see The fitting method adopted in this study is described in details in Section Materials and methods.
To verify the reliability of the proposed fitting method, we performed numerical • The absorption coefficients for the two layers were obtained by adding to ∆µa up (λ,Τ) and ∆µa down (λ,Τ) the baseline absorption coefficients µa0(λ), as obtained in Supplementary Section S2.
• Finally, the time courses for the changes in optical density at short and long source detector distance were calculated from Rshort(λ,Τ) and Rlong(λ,Τ): ΔODshort ( were converted to ΔOD, which is valid on the assumption that the head is a homogeneous medium.