The default mode network is a task negative functional network in the brain. Variations in the strength of this network have been associated to many neurological diseases, such as dementias, schizophrenia and autism [1], though few studies have attempted to elucidate the effect of brain tumours on the default mode network. The few studies that have been conducted have found that tumours cause a decrease in DMN connectivity, but they have not attempted to differentiate whether certain tumour types have a larger effect on the DMN [2-4]. The goal of this study was to investigate the effect of tumour histological type and location on its effect on the default mode network.

Data acquisition

73 patients afflicted with brain tumours and 23 healthy aging patients were recruited for this study. For each patient, T1 MPRAGE anatomical scans (1mm isotropic) and resting state fMRI scans (EPI sequence, resolution of 3.75x3.75x3 mm with TR/TE of 3.864/0.04 s) were taken with a Siemens 1.5 T MR scanner. Diagnosis of the tumour afflicted patients was made by histology.

Pre-processing

fMRI data was pre-processed in accordance to current norms for resting-state connectivity studies [5].Pre-processing was conducted with the AFNI software package [6]. Briefly, functional data was despiked, motion corrected, coregistered to the high resolution skull stripped anatomical scan and blurred with a 4mm FWHM Gaussian kernel. Both functional and anatomical data was then warped to an MNI template to conduct multi-subject statistical testing.

Connectivity analysis

Functional connectivity was probed with two commonly used modalities: seed-based analysis (SBA) and independent component analysis (ICA). Both modalities were used in order to benefit from the advantages associated to each. For SBA, 5mm radius seeds were placed at 6 locations commonly known as being DMN nodes: the posterior cingulate cortex, the ventromedial prefrontal cortex, the left and right angular gyri and the left and right parahippocampi. Average time series were extracted from these spheres and the correlation coefficient of the time series of each DMN node with every other node was computed. This produced 15 unique correlation values, which were averaged together to produce a measure of overall connectivity for each subject. ANOVA tests were then done to find whether certain tumour types or tumour locations had significantly higher effects on the DMN. Furthermore, ICA with a temporal concatenation approach was performed with FSL’s MELODIC software package [7]. This method involves temporally concatenating all functional datasets (of all patients) and then performing ICA to extract a group wide DMN map. Then, subject DMN maps were backwards constructed using the recently developed dual regression method [8]. Briefly, the group-wide DMN spatial map was used as a spatial regressor to produce a DMN related “timeseries” for each patient; these time series were then used as a temporal regressor in a second regression analysis to extract DMN spatial maps for each subject. Then, non-parametric permutation tests were performed with 5000 permutations in order to test for significant differences between different groups of subjects based on tumour location. The multiple comparisons problem in this case was controlled with a cluster mass method derived from random field theory.

The results are summarized in figure 4. With SBA and ANOVA testing, we found no significant differences between any two subdivisions of the tumour patient group. Tumour location, compartment (intra-axial vs extra-axial), histological type (glioma, meningioma, metastasis, lymphoma, kyste) and aggressiveness (infiltrative vs non infiltrative) were not found to be significant predictors of the tumour’s impact on the DMN. However, the lack of statistically significant results may be explained by the sensitivity of correlation analyses to the brain distorting effect of tumors. In fact, improper seed placement has been found to be an important confound in such correlation analyses [9]; anatomy distortion by massive tumors may have caused such improper seed placement.

The results of the ICA coupled with dual regression and non-parametric permutation tests are summarized in figure 1, 2 and 3. These tests found that tumours located in the left occipito-parietal lobe and the left temporal lobe caused statistically significant decreases in DMN connectivity, whereas tumors situated in the right sided counter parts of these regions did not. Moreover, tumors in the cerebellum also caused decreases in DMN connectivity.

Tumours in the frontal lobe caused nearly statistically significant decreases in DMN connectivity. These results, coupled with other studies done on the effect of tumours on the DMN, may lead to brain connectivity based prognostic scores which might help clinicians in evaluating the appropriateness of certain intervention in patients with brain tumours.