Participants

Fifteen patients with first-episode schizophrenia and 15 healthy volunteers of comparable age, gender, handedness and parental educational level (Table 1), rated as in our previous study (Reference Jensen, Al-Semaan and DrostJensen et al, 2002b ), participated in the study after giving informed, written consent according to the guidelines of the Health Sciences Research Ethics Board at the University of Western Ontario. The DSM–IV diagnosis of schizophrenia (American Psychiatric Association, 1994) was established using the Structured Clinical Interview for DSM–IV (SCID; Reference First, Spitzer and GibbonFirst et al, 1997), administered by a psychiatrist (P.C.W.). Twelve cases were classified as having paranoid schizophrenia and three as having undifferentiated schizophrenia. Ten patients were neuroleptic-naïve, but three of the ten had received a benzodiazepine for sedation at the time of the scan and another three had been previously exposed to a selective serotonin reuptake inhibitor. Of the five patients receiving neuroleptic medication at the time of the scan, four were taking atypical antipsychotic agents and one a conventional neuroleptic, with two patients having had prior exposure to a selective serotonin reuptake inhibitor. The estimated mean duration of exposure to antipsychotic medication at the time of scanning for the five patients was 0.4 (s.d.=0.4) years. Ratings of symptoms on the Scale for the Assessment of Negative Symptoms (SANS; Reference AndreasenAndreasen, 1984a ) and the Scale for the Assessment of Positive Symptoms (SAPS; Reference AndreasenAndreasen, 1984b ) were 39 (s.d.=11) and 28 (s.d.=17), respectively. Mean duration of untreated psychosis since the onset of positive symptoms was 0.7 (s.d.=0.7) years (Reference Norman and MallaNorman & Malla, 2001). The healthy volunteers were recruited from the community by advertisement and were assessed by a psychiatrist (P.C.W.) using the SCID to exclude psychiatric diagnoses or a history of psychiatric disorders in a first- or second-degree relative. No patient or volunteer had a history of head injury or of drug or alcohol misuse in the 2 years before the scan, or had any current serious medical illness.

Table 1

Characteristics of the study participants

	Control group	Schizophrenia group
Gender (n)
Male	13	13
Female	2	2
Right-handed (n)	15	14
Age, years (mean (s.d.))	22.1 (3.0)	22.5 (3.4)
Parental education1 (mean (s.d.))	3 (1)	3 (1)
Length of illness, years (mean (s.d.))		0.7 (0.7)
Total SANS score (mean (s.d.))		39 (11)
Total SAPS score (mean (s.d.))		28 (17)

Imaging technique

All experiments were performed on a 4.0 Tesla Varian/Siemens/Unity Inova whole-body magnetic resonance scanner (Varian, Palo Alto, California, USA: Siemens, Erlangen, Germany) operating at 170.3 MHz. A single-tuned, proton quadrature hybrid-birdcage volume head coil was used for shimming and imaging. The individuals to be examined were positioned supine with their heads secured in a Plexiglas cradle and the global water resonance was manually shimmed. A two-dimensional fast low-angle shot imaging sequence acquired anatomical, sagittal and coronal images of the brain for ³¹P voxel positioning (repetition time 11 ms, echo time 6 ms, 256 phase-encodes, 0.93 mm × 0.93 mm pixels). Thirty-two T ₁-weighted transverse images were acquired on the brain using a magnetisation-prepared (adiabatic inversion pulse 0.5 s prior to data collection) three-dimensional fast low angle shot imaging sequence (repetition time 11 ms, echo time 6.2 ms, total acquisition time <3 min) with a nominal slice thickness of 4.375 mm and 1.09 × 1.09 mm in-plane resolution. These images were used to calculate grey matter, white matter and cerebrospinal fluid ratios in the ³¹P chemical shift imaging volumes.

Image segmentation and partial volume estimation

The transverse image data-set for each brain examined was first separated into three distinct binary tissue maps – grey matter, white matter and cerebrospinal fluid – by thresholding the original high-contrast image set. Two thresholding values were obtained from the central slice of each image set, which best displayed the ventricles, corpus callosum and grey matter of the anterior cingulate. The thresholding coefficients were then varied until each segmented image type matched the corresponding anatomy in the original reference image. The thresholding values differed between the individuals examined since image signal intensity depended on receiver gain and coil loading. Extraneous tissue with signal intensities similar to neural tissue or cerebrospinal fluid, such as skin and subcutaneous lipid, was removed from the segmented images using commercial software (Analyze 4.0 for PC; Biomedical Imaging Resources, Mayo Clinic) to avoid it being counted as neuronal tissue in the final volumetric analysis. To obtain the tissue ratio contribution for each ³¹P region of interest, the three segmented image datasets were convolved with the calculated three-dimensional point spread function of the ³¹P chemical shift imaging acquisition (Reference Jensen, Drost and MenonJensen et al, 2002a ). The partial volume contribution for each tissue type was then expressed as a percentage of total tissue contribution for each region of interest (see Table 3).

Table 3

Voxel tissue partial volume estimates for brain regions predicted to show differences in phosphomonoester and phosphodiester levels between groups

Region	Grey matter partial volume fractions (%)		White matter partial volume fractions (%)		Cerebrospinal fluid partial volume fractions (%)
	Schizophrenia group mean (s.d.)	Control group mean (s.d.)	Schizophrenia group mean (s.d.)	Control group mean (s.d.)	Schizophrenia group mean (s.d.)	Control group mean (s.d.)
Anterior cingulate	43 (11)	44 (16)	39 (19)	52 (16)	17 (17)	4 (3)
Left thalamus	11 (10)	5 (3)	87 (11)	93 (4)	2 (2)	1 (1)
Right thalamus	11 (11)	5 (3)	87 (12)	93 (4)	2 (2)	2 (1)
Left cerebellum	26 (18)	29 (20)	72 (21)	70 (20)	3 (4)	1 (1)
Right cerebellum	28 (19)	35 (21)	69 (20)	64 (21)	2 (2)	2 (2)
Left hippocampus	23 (18)	19 (16)	74 (23)	79 (16)	4 (7)	2 (1)
Right hippocampus	21 (17)	22 (17)	77 (19)	77 (17)	2 (4)	1 (1)
Left prefrontal	38 (12)	35 (8)	52 (20)	62 (10)	10 (18)	3 (4)
Right prefrontal	42 (21)	33 (9)	45 (27)	65 (9)	13 (21)	2 (3)

In vivo ³¹P–MRS

Once the ¹H shimming and imaging were completed, the ¹H head coil was manually replaced with a single-tuned, ³¹P quadrature, 14 cm long birdcage head coil, with the person to be examined still in place in the Plexiglas cradle inside the magnet. Previous phantom experiments verified that manual coil replacement did not affect the shim. A ³¹P external reference standard – methylene diphosphonic acid, 270 mol/l, T ₁ at 4.0 Tesla about 6 s – was fastened to the Plexiglas head cradle just to the left of the person's head. The 1.5 cm diameter reference tube was positioned so its length ran axially, spanning the sensitive region of the ³¹P radio frequency coil. Transmit/receive frequency was centred on phosphocreatine, as measured with a global free induction decay. Tip angle, optimised for the phosphomonoester/phosphodiester resonances at a 500 ms repetition time, was 32°. In vivo ³¹P–MRS used an optimised three-dimensional chemical shift imaging sequence (Reference Jensen, Drost and MenonJensen et al, 2002a ): repetition time 500 ms; sampling bandwidth ± 2 kHz; complex points 1600; readout duration 400 ms; pre-pulses 20; receiver gain 94 dB; pre-acquisition delay 1.905 ms; field of view (x,y,z) 280 mm; nominal volume 5.4 cm³; maximum phase-encode matrix dimension (x,y,z) 14 × 14 × 14 (zero-filled out to 16 × 16 × 16 prior to reconstruction). The three-dimensional chemical shift imaging sequence used a reduced phase-encoding scheme allowing for the inclusion of spherically bound, reduced point, weighted k-space acquisition, providing approximately 40% more signal-to-noise for a given scan time and effective voxel volume over conventional methods (Reference Jensen, Al-Semaan and DrostJensen et al, 2002b ).

In vivo post-processing and spectral analysis

All in vivo chemical shift imaging/image data were processed and viewed using Varian Nuclear Magnetic Resonance software (version 6.1B) and software designed and written on site. Prior to fast Fourier transform reconstruction to spatially resolve the chemical shift imaging spectra, the collected k-space data were centred in a 16 × 16 × 16 cubic matrix. Each time-domain free induction decay was then zero-filled out to 2048 complex points and left-shifted 5 points to remove residual bone and rigid membrane signal.

Using the ¹H images, the three-dimensional chemical shift imaging data grid was shifted in the x, y and z dimensions in order to position each voxel in the appropriate ³¹P region of interest. For all brains examined, the Brodmann area and Talairach coordinates for voxel centres (x,y,z) were evaluated for every ³¹P region of interest using statistical parametric mapping (Table 2). Partial overlap with adjacent structures was unavoidable in many of the regions owing to voxel volume and shape (Fig. 1).

Table 2

Brodmann areas and Talairach coordinates for regions of interest

Region	Brodmann area	Talairach coordinates1
		x	y	z
Anterior cingulate	32	0	44	2
Posterior cingulate	23	0	— 48	26
Left prefrontal	10	— 22	42	2
Right prefrontal	10	22	42	2
Left thalamus		— 9	— 11	5
Right thalamus		9	— 11	5
Left parieto-occipital	39	— 31	— 56	26
Right parieto-occipital	39	31	— 56	26
Left cerebellum		— 20	— 61	— 26
Right cerebellum		20	— 61	— 26
Left hippocampus		— 24	— 29	— 12
Right hippocampus		24	— 29	— 12

Fig. 1

Voxel positions: regions of interest displayed on transverse localiser images. (A) Prefrontal, 1; anterior cingulate, 2; parieto-occipital, 3; posterior cingulate, 4. (B) Thalamus, 5; hippocampus, 6. (C) Cerebellum, 7. Corresponding voxel positions are displayed on sagittal (D) and coronal images (E, F). Effective voxels shown represent actual size with respect to images and are 3.1 cm in diameter (15 cm³ volume).

All in vivo spectra were fitted in the time domain using a non-linear, iterative fitting program developed on site (Fig. 2). The fitting routine is based on the Marquardt–Levenberg algorithm, using prior spectral knowledge for the relative amplitudes, line widths, line shapes, peak positions and J-coupling constants to model the in vivo ³¹P brain spectrum (Reference Jensen, Drost and MenonJensen et al, 2002a ). Since it was positioned along the inner face of the coil, the reference standard was subject to different radio frequency power from that used for the brain, and was volume-bound by the x and y tube dimensions (1.5 cm diameter), both requiring correction. Therefore, the methylene diphosphonic acid reference area, modelled with a triplet structure, was corrected for T ₁ saturation, radio frequency coil sensitivity and volume. Each ³¹P metabolite area was also T ₁-corrected and then normalised to the methylene diphosphonic acid signal to obtain absolute millimolar values per unit of wet brain-tissue volume (mol/l) for each ³¹P metabolite (Reference Jensen, Drost and MenonJensen et al, 2002a ). Human in vivo ³¹P high-energy phosphate metabolite T ₁ values at 4.0 Tesla for inorganic phosphate, phosphocreatine and adenosine triphosphate of 3.06 s, 2.98 s and 1.07 s were used for the T ₁ correction, as well as T ₁ values of 2.68 s, 1.5 s, 3.92 s, 3.99 s and 2.31 s for phosphoethanolamine, phosphocholine, glycerophosphoethanolamine, glycerophosphocholine and membrane phospholipids, respectively (Reference Jensen, Drost and MenonJensen et al, 2002a ). The fitted metabolite amplitudes are not T ₂-weighted since the fitting algorithm extrapolates back to time zero. Each ³¹P metabolite value was finally corrected for cerebrospinal fluid partial volume fraction assuming phosphate concentration in cerebrospinal fluid to be negligible. There was no correction for the ratio of grey to white matter.

Fig. 2

In vivo ³¹P brain spectrum from a 15 cm³ effective voxel in the left thalamus of a healthy volunteer at 4 Tesla. Raw data are displayed with modelled fit and residual; 10 Hz exponential filtering has been applied for display. ATP, adenosine triphosphate; GPCho, glycerophosphocholine; GPEtn, glycerophosphoethanolamine; PCho, phosphocholine; PCr, phosphocreatine; PDE, phosphodiester; PEtn, phosphoethanolamine; P_i,inorganic phosphate; PME, phosphomonoester; ppm, parts per million; MP, membrane phospholipids.

Statistics

Statistical calculations were performed using the Statistical Package for the Social Sciences, version 10.0 for Windows. A two-tailed t-test was used to test for significant differences in age, parental education, grey matter, white matter and cerebrospinal fluid partial volumes between groups (P < 0.05). In the same fashion as in our previous study (Reference Jensen, Al-Semaan and DrostJensen et al, 2002b ), statistical treatment of all ten ³¹P metabolite ratios for each region first involved a one-factor multivariate analysis of variance (MANOVA) with a between-subjects factor of group. Pursuant to a significant multivariate test, univariate F-tests were used to examine group differences, with α set at P < 0.05. In each region where significant group differences in ³¹P metabolites were found, Pearson correlations (P < 0.01) were computed within the patient group between ³¹P metabolite levels and each of the following variables: SANS and SAPS scores, age, parental education and duration of untreated psychosis.

Our application of the region-wise MANOVA represents an optimal compromise between power and type I error protection. Expanding the analysis to include region as a within-subject factor, resulting in the split-plot factorial MANOVA applied to the vector of the metabolites, unnecessarily complicates the analysis when there are a priori reasons to select the targeted regions (Reference StevensStevens, 1996). The issue of following up a significant MANOVA test, and whether or not to apply Bonferroni correction, has been addressed in the literature (Reference Hummel and SligoHummel & Sligo, 1971; Reference StevensStevens, 1996), justifying the statistical treatment of our data.

Implications of altered phosphodiester levels

Phospholipase A1 and A2 activity produces the phosphodiester metabolites glycerophosphocholine and glycerophosphoethanolamine. Phosphodiesterase, a catalytic enzyme responsible for the breakdown of phospholipid membrane, also produces glycerophosphocholine and glycerophosphoethanolamine. A decrease in the activity of this enzyme, and/or an increase in phospholipase A1 and A2 activity, can decrease phosphodiester levels in the brain, and may be involved in the development of schizophrenia (Reference Horrobin, Glen and VaddadiHorrobin et al, 1994). However, the increase in glycerophosphocholine seen in the anterior cingulate may suggest a reversal of this mechanism, leading to increased membrane breakdown during the acute stages of the illness and eventual neuronal loss in this region as the illness progresses. This is supported by our earlier study in patients with chronic illness at 4.0 Tesla (Reference Jensen, Al-Semaan and DrostJensen et al, 2002b ), which found decreased glycerophosphoethanolamine levels in the anterior cingulate. Interestingly, no difference in any phospholipid metabolite was detected between groups in either the left or the right prefrontal cortex between the first-episode schizophrenia group and the controls. This contrasts with the results of other studies, which report increased prefrontal phosphodiester levels in first-episode but not chronic illness (Reference Williamson, Drost, Glen and HorrobinWilliamson & Drost, 1999) and decreased phosphodiester levels in drug-naïve (not first-episode) patients in the left frontal lobe (Reference Yacubian, de Castro and OmettoYacubian et al, 2002). However, these earlier studies were conducted at lower field strengths (<2 Tesla), where (unlike 4.0 Tesla imaging) the broad phosphodiester resonance is dominant in the human ³¹P brain spectrum. This fact, combined with different voxel volumes and analytical techniques might possibly contribute to our different findings regarding the prefrontal cortex. Our ³¹P–MRS studies of the brains of patients with chronic and first-episode schizophrenia at 4.0 Tesla suggest that the prefrontal cortex may manifest abnormalities in phospholipid metabolism only later in the chronic stages of illness, suggestive of eventual neurodegeneration. Indeed, if neurodegeneration is the primary course of evolution in this illness, then one could speculate that early intervention with neuroprotective agents could halt – or at least slow down – the onset and development of schizophrenic symptoms, lessening the negative social, physical and economic impact of this disease.

Other metabolites

High-energy phosphate changes seen in this study of first-episode schizophrenia occurred in the anterior cingulate, with increased concentrations of inorganic phosphate, phosphocreatine and adenosine triphosphate. Interpretation of this finding is somewhat difficult as adenosine triphosphate levels should remain constant, regardless of the metabolic demands placed on the neural tissue. A decreased metabolic rate might explain the increased phosphocreatine level in this region, since the demand for replenishing adenosine triphosphate levels is reduced and phosphocreatine stores are not being drawn upon, hence the increased levels. One explanation for the increased adenosine triphosphate level in this region might be that there is a mitochondrial dysfunction in these neurons, since the mitochondria are responsible for producing the cell's supply of adenosine triphosphate. Mitochondrial dysfunction in neuropsychiatric disorders has been alluded to in other illnesses yielding changes in adenosine triphosphate levels in the brain, such as depression (Reference Volz, Rzanny and RiehemannVolz et al, 1998). Since schizophrenia has a strong genetic component, mitochondrial dysfunction could at least partially explain the energetic anomalies seen in this and other ³¹P–MRS studies of schizophrenia (Reference Jensen, Al-Semaan and DrostJensen et al, 2002b ). However, this hypothesis has yet to be tested.

Effect of medication

The effects of neuroleptic and anxiolytic medications on brain phosphorus metabolites warrant further discussion. Two human studies in vivo examining the effect of neuroleptic administration on phosphorus metabolites suggest that phosphodiester levels are decreased in the temporal lobe (Reference Fukuzako, Fukuzako and KodamaFukuzako et al, 1999) and increased in the frontal lobe (Reference Volz, Rössger and RiehemannVolz et al, 1999) with neuroleptic medication. In this study, we observed no differences in phosphodiester levels in either of these regions between groups. However, it cannot be ruled out that phosphodiester level differences in these regions do actually exist, but are masked by the effects of neuroleptics. There are very few studies examining the effects of benzodiazepine medications on brain phosphorus levels. A study by Miranda et al (Reference Miranda, Ceckler and Guillet1990) found that phosphocreatine utilisation is altered in the adult rat brain only after in utero exposure to benzodiazepine ligands. Extrapolating this result to the humans investigated in this study, one can only speculate that phosphocreatine levels might be affected by maternal use of benzodiazepines during pregnancy. However, it is clear that more human research is needed in this area to test this hypothesis.

Spectroscopy limitations

The repetition time used for our ³¹P–MRS acquisition (500 ms) was very short compared with the measured T ₁ values of the phosphorus metabolites. The resulting ³¹P spectra are therefore T ₁-weighted despite radio frequency pulsing at the Ernst angle for a T ₁ of 3 s (the average T ₁ of phosphomonoester, phosphodiester and phosphocreatine metabolites). It is possible that our observed changes in metabolite levels between groups might partly result from different T ₁ values in the affected regions, aside from concentration differences. This is unlikely, however, since T ₁ values in people with schizophrenia would have to differ by up to several hundred per cent in order to equalise the metabolite ratios between the groups, and there is no supporting evidence from other studies suggesting that brain phosphorus metabolite T ₁ values are altered in schizophrenia.