Multidimensional Scaling of Drug Effects in CorticoStriatoThalamic Circuits

1
Multidimensional scaling of drug effects in
cortico-striato-thalamic circuits
D Welchew, G Honey, J Suckling, F Zelaya, C
Long, C Routledge, V Ng, P Fletcher, J
Brown, S Williams, Ed Bullmore
Cambridge University Department of Psychiatry,
Institute of Psychiatry (Kings College),
GlaxoSmithKline Clinical Research Unit
(Cambridge)
E-mail dew22_at_cam.ac.uk
INTRODUCTION
Univariate and Multivariate Analysis

• Multidimensional scaling (MDS) techniques may be
  used in functional magnetic resonance imaging
  (fMRI) to define sub-systems of functionally
  connected brain regions identified by established
  brain activation mapping techniques1.
• Here we apply a permutation test based on
  Procrustes analysis of configurations of regions
  to investigate the effects of neurotransmitter
  systems on functional connectivity.

METHODS

• Above left Regions activated by univariate
  analysis.
• Above right Three dimensional configurations of
  points under two-way MDS for each treatment
  condition. Note in particular the changing
  position of the thalamus relative to other
  subcortical structures.

Imaging and Subjects

• Gradient echo echo-planar imaging at 1.5 Tesla
  was used to study 12 right-handed, elderly
  volunteers (5 male mean age 72.4 years)
  performing a paired-associate learning task.
• The subjects were treated with methylphenidate,
  sulpiride or placebo on each of three separate
  occasions. Treatment order was randomised across
  subjects.
• We used a periodic ABAC design. A (24 seconds)
  crosshair fixation. B (24 seconds)
  identification of the location of a probe item in
  a previously presented array of two items. C (24
  seconds) as B with an array of four items. Probe
  location remained constant throughout the
  experiment.
• Generic brain activation maps of regions
  responsive to the load manipulation were
  constructed from the individual images and used
  to identify seven regions of interest.

Procrustes Analysis

• Above left Summary of permutation distributions
  for the comparison between the sulpiride and
  placebo conditions.
• Above centre Permutation distribution for the
  thalamus. The position of this region was
  significantly altered in relation to the rest of
  the areas studied (plt0.05 corrected).
• Above right Permutation distribution for sum
  squared residual difference between
  configurations. This histogram demonstrates
  marked bimodality (see below).

Statistics

• Time series for nine randomly selected voxels
  within each ROI and for each subject were
  averaged and then mean-corrected, and used as the
  basis for two and three-way multidimensional
  scaling2 for each drug treatment.
• Several dimensions were identified using MDS, and
  each drug was compared to the placebo by
  generating a configuration of the ROIs in these
  dimensions.
• For each comparison, treatments were randomly
  reallocated to each subject and pairs of
  configurations generated by applying generalised
  Procrustes analysis to translate, rotate and
  scale one onto the other3.
• The residual mean squared difference between
  points was used to construct distributions
  against which the observed points could be
  compared.
• Three-way MDS was used to examine the
  heterogeneity of subjects within each treatment
  group.
• Individual differences MDS using the ALSCAL
  algorithm4 was used to generate a Weirdness
  statistic for each subject, and these were
  compared between groups using Students t-test.

Weirdness

• No significant difference was found in Weirdness
  between groups.
• Between - subject effects dominate any effect of
  medication on heterogeneity.

CONCLUSIONS

• Multidimensional scaling provides several useful
  methods for examining functional connectivity in
  imaging data.
• We have demonstrated techniques for assessing the
  statistical significance of differences found
  between groups and treatments, allowing these
  techniques to be used in a confirmatory as well
  as exploratory way to characterise both group
  differences and individual variance as a result
  of pharmacological intervention.
• Although these results do not conclusively
  demonstrate an overall difference in
  configuration under different pharmacological
  challenges, they suggest that the relationship
  between particular subsystems may be altered by
  dopaminergic modulation particularly the
  striato-thalamo-mesencephalic regions. This
  agrees with findings from correlation analysis.
• These results show the need to sample a wide
  range of ROIs where possible for this technique,
  to allow more optimal numbers of dimensions to be
  found. This improves power by reducing bimodality
  in the permutation distributions.

RESULTS

• One bilateral and five right-sided ROIs were
  identified Dorsolateral prefrontal cortex
  (bilateral, RDLPFC/LDLPFC, Talairach co-ordinates
  mm 24,24,28 and 18,30,28), ventral prefrontal
  cortex (VPFC, 25,33,-4), ventral midbrain (VM
  22,-30,4), caudate nucleus (CD, 15,18,1), putamen
  (PUTM, 29,-1,4) and ventro-posterior thalamus
  (THAL, 17,33,4).
• Multidimensional scaling identified
  three-dimensional solutions providing optimal
  reduction in stress, separating regions on the
  basis of fronto-striatal vs. other,
  dorsolateral PFC vs. other, and subcortical
  vs. cortical.

REFERENCES

• Brammer M.J., Bullmore E.T. et al. Magnetic
  Resonance Imaging 1997, 15763-770
• Torgerson W. S. Psychometrika 1952, 17401-419
• Welchew D.E., Honey G.H. et al. Neuroimage
  Submitted
• Takane Y., Young F.W. and de Leeuw J.
  Psychometrika 1977, 427-67

DEW was supported by an MRC PhD studentship.