OUP user menu

Selective neuronal loss in rescued penumbra relates to initial hypoperfusion

J. V. Guadagno, P. S. Jones, F. I. Aigbirhio, D. Wang, T. D. Fryer, D. J. Day, N. Antoun, I. Nimmo-Smith, E. A. Warburton, J. C. Baron
DOI: http://dx.doi.org/10.1093/brain/awn175 2666-2678 First published online: 4 August 2008


Selective neuronal loss (SNL) in the rescued penumbra could account for suboptimal clinical recovery despite effective early reperfusion. Previous studies of SNL used single-photon emission tomography (SPECT), did not account for potential volume loss secondary to collapse of the infarct cavity, and failed to show a relationship with initial hypoperfusion. Here, we obtained acute-stage computerized tomography (CT) perfusion and follow-up quantitative 11C-flumazenil (FMZ)-PET to map SNL in the non-infarcted tissue and assess its relationship with acute-stage hypoperfusion. We prospectively recruited seven patients with evidence of (i) acute (<6 h) extensive middle cerebral artery territory ischaemia based on clinical deficit (National Institutes of Health stroke scale, NIHSS score range: 8–23) and CT Perfusion (CTp) findings and (ii) early recanalization (spontaneous or following thrombolysis) based on spectacular clinical recovery (ΔNIHSS 6 at 24 h), good clinical outcome (NIHSS 5) and small final infarct (6/7 subcortical) on late-stage MRI. Ten age-matched controls were also studied. FMZ image analysis took into account potential post-stroke volume loss. Across patients, clusters of significantly reduced FMZ binding were more prevalent and extensive in the non-infarcted middle cerebral artery cortical areas than in the non-affected hemisphere (P = 0.028, Wilcoxon sign rank test). Voxel-based between-group comparisons revealed several large clusters of significantly reduced FMZ binding in the affected peri-insular, superior temporal and prefrontal cortices (FDR P < 0.05), as compared with no cluster on the unaffected side. Finally, comparing CTp and PET data revealed a significant negative correlation between FMZ binding and initial hypoperfusion. Applying correction for volume loss did not substantially alter the significance of these results. Although based on a small patient sample sometimes studied late after the index stroke, and as such preliminary, our results establish the presence and distribution of FMZ binding loss in ultimately non-infarcted brain areas after stroke. In addition, the data suggest that this binding loss is proportional to initial hypoperfusion, in keeping with the hypothesis that the rescued penumbra is affected by SNL. Although its clinical counterparts remain uncertain, it is tempting to speculate that peri-infarct SNL could represent a new therapeutic target.

  • stroke
  • cerebral ischaemia
  • 11C-flumazenil
  • benzodiazepine receptor
  • cerebral infarction
  • positron emission tomography
  • CT perfusion


The fate of the reperfused ischaemic penumbra is currently attracting considerable interest (Kidwell et al., 2002; Baron, 2005). For instance, in both animals and man, the acute diffusion-weighted imaging (DWI) lesion can be reverted following reperfusion (Kidwell et al., 2000; Ringer et al., 2001) but may partially reappear, suggesting that although rescued from pan-necrosis this tissue may have suffered a milder degree of damage (Li et al., 2000; Kidwell et al., 2002). Studies in rat models have consistently documented selective neuronal loss (SNL) in ischaemic tissue rescued from infarction by early reperfusion (Garcia et al., 1996; Sicard et al., 2006) while subtler neuronal damage is found if arterial occlusion is of very short duration (Garcia et al., 1997; Li et al., 2000; Ringer et al., 2001). Because damage to cortical neurons is likely to hamper restoration of function (Weiller et al., 1993; Baron, 2005), documenting its occurrence and distribution in man is an important goal.

11C-Flumazenil (FMZ), an antagonist of the central benzodiazepine receptor (cBZR; a component of the ubiquitous GABAa complex), is the reference in vivo radioligand for quantitative studies of neuronal density (Herholz et al., 2004), and has been successfully used to map the ischaemic core in acute stroke in both man and animal models (Sette et al., 1993; Heiss et al., 1997, 1998, 2001). FMZ should, therefore, be an ideal tool to map SNL in the reperfused penumbra. However, both clinical studies that addressed this issue (Nakagawara et al., 1997; Saur et al., 2006) used 123I-Iomazenil (IMZ), a single-photon emission tomography (SPECT) analogue of FMZ that has less favourable kinetic properties and whose binding is strongly influenced by initial delivery (Videbaek et al., 1993; Nakagawara et al., 1997; Saur et al., 2006). Furthermore, SPECT has poor intrinsic spatial resolution and is not suited for accurate quantification. Despite these limitations, both studies reported reduced IMZ in the reperfused but structurally normal cortex. However, Saur et al. (2006) assessed IMZ uptake rather than specific binding, and did so in the early subacute stage where the blood–brain barrier may be disrupted; while in the paper by Nagakawara et al. (1997) the IMZ, perfusion and follow-up structural scans were not co-registered and the potential confound of cortical volume loss due to secondary collapse of infarct cavity was not considered.

A key point when addressing SNL after stroke is whether its expected relationship with the degree of acute-stage hypoperfusion based on early animal work (Heiss and Rosner, 1983) is recovered. Nagakawara et al. (1997) did not assess this relationship, while Saur et al. (2006) reported only a weak non-significant trend in the expected direction, which may have resulted from their use of large regions of interest (ROIs) on top of the limitations already mentioned.

In this study, we combined acute-stage CT perfusion (CTp) and subacute-to-chronic quantitative PET to map FMZ binding and assess its relationship with initial hypoperfusion. We predict that FMZ binding will be reduced in the non-infarcted cortex of the affected, as compared with the non-affected, cerebral hemisphere and that the loss of FMZ binding will be linearly proportional to initial hypoperfusion. This design mirrors the penumbra model, where infarction assessed days to weeks after the ischaemic insult is related to acute-stage perfusion; indeed, tissue damage caused by acute ischaemia is clearer in the subacute-to-chronic rather than the acute stage (Astrup et al., 1981; Jones et al., 1981; Garcia et al., 1983). The same approach has been used in animal studies of SNL (Mies et al., 1983; Garcia et al., 1996; Sicard et al., 2006). Also, it seems advisable to delay scanning to the subacute or chronic stage when using in vivo ligand binding as a surrogate of neuronal loss because cell death secondary to reperfusion injury may be delayed (Dirnagl et al., 1999), and the loss of receptor membrane protein may itself be delayed relative to cell death. As for when is the best time, previous work has shown the loss of cBZR binding in the rescued penumbra is stable from 2 to 3 weeks onwards (Nakagawara et al., 1997). Our design was therefore to look at FMZ binding after 3 weeks, with the proviso that recurrence of transient ischaemic attack (TIA) or stroke following the index event was a cause for exclusion. Finally, because of the risk of brain volume loss at this late stage, partial volume effects were also controlled for in the analysis.

Material and Methods



To optimize the chances of detecting SNL, we prospectively recruited patients admitted within 6 h of onset with (i) clinical evidence of extensive middle cerebral artery (MCA) territory stroke (admission NIHSS ≥ 8); (ii) remarkable early recovery (ΔNIHSS ≥ 6 at 24 h), either spontaneously or following thrombolysis; (iii) excellent outcome (NIHSS ≤ 5 on the day of PET) and (iv) small final infarct on structural MRI obtained on the same day as the PET. Exclusion criteria were: (i) haemorrhage on admission CT; (ii) stroke recurrence or TIA after the index stroke; (iii) previous history of any intracranial pathology or cognitive impairment; (iv) major medical co-morbidity including alcohol abuse (i.e. >4 units/day for men and 3 units/day for women) and (v) patients on long-term prior, or any current benzodiazepine. All recruited patients were prospectively scored with the NIHSS acutely, 24 h post-ictus and on the day of PET scanning. To assess the relationship between initial hypoperfusion and subsequent SNL, CTp was obtained on admission, immediately following the plain CT. The local ethics committee approved the study and written informed consent was obtained.


Ten age-matched healthy controls (seven females/three males, mean age 57, range 48–71) were studied. They had no history of any medical condition, were not alcohol abusers and were taking no medication.


A structural MRI was obtained in all subjects on the day of PET scanning. The protocol included high-resolution 3D volume T1-weighted spoiled-gradient (SPGR; voxel size 0.70 × 0.86 × 1.00 mm, 256 × 256 × 256 matrix), T2-weighted (voxel size 0.66 × 0.70 × 5.00 mm; slice thickness 4 mm; 256 × 512 × 27 matrix) and fluid-attenuation inversion-recovery (FLAIR; voxel size 0.66 × 0.70 × 5.00 mm; slice thickness 4 mm, 256 × 512 × 27 matrix) sequences, acquired on a 3T whole body magnet (Med-spec s300, Bruker, Ettlingen, Germany).


An FMZ-PET study was obtained in all healthy controls and patients ≥20 days after the index stroke, using a General Electric Advance PET Scanner (Milwaukee, WI; effective isotropic resolution: 6.8 mm). A 68Ge transmission scan was first carried out to correct for attenuation. FMZ was labelled with 11C using a novel methylation process (Cleij et al., 2007), providing high specific activities (370–550 GBq/μmol). The compound was injected intravenously as a bolus (275–480 MBq) followed by 60-min acquisition (52 dynamic frames: 5 s × 18, 15 s × 6, 30 s × 10, 60 s × 7, 150 s × 4 and 300 s × 7) of 35 slices with voxel dimensions of 2.34 × 2.34 × 4.25 mm (x, y, z).

The arterial input function (AIF) was determined in control subjects from 16 samples obtained at predetermined time points via a radial cannula. A further six samples were taken throughout for assay of radiolabelled metabolites. Arterial cannulation was not considered essential in patients—and was contraindicated in several—because assessing receptor availability using FMZ-binding potential (BPND) maps with the pons as a reference tissue (Lucignani et al., 2004) has been validated against compartmental modelling using the AIF (Klumpers et al., 2007). To confirm the validity of BPND maps in our laboratory, we obtained both BPND maps and gold-standard distribution volume (VT) maps in our sample of healthy controls for comparison.

CT perfusion

Immediately following the admission plain CT, CTp was carried out on a helical CT scanner (Sensation 4 model, 120 pKV, 258 mA; Siemens GmbH, Erlangen, Germany), using 50 ml of iopamidol injected into the cubital vein (flow rate 8 ml/s) using an injection pump. Sequential acquisition (one image/s for 40 s) was obtained for two axial slices (voxel dimensions 0.42 × 0.42 × 10 mm) centred to include the basal ganglia and the plane above.

Image processing

FMZ parametric maps

First, the FMZ added image from all frames was co-registered to the structural SPGR using statistical parametric mapping (SPM2), and the transformation matrix applied to all PET frames and post-processed parametric maps. Second, from the dynamic FMZ data set and the metabolite-corrected AIF, voxel-by-voxel maps of VT were generated (in control subjects), based on the classic 2-compartment, 2-parameter model (Frey et al., 1991; Koeppe et al., 1991), using the Logan plot implemented in PET Modelling (PMOD, Zurich, Switzerland). Finally, BPND maps were generated (in all subjects) using the simplified reference tissue model (SRTM) (Gunn et al., 1997). Circular ROIs, five voxels in diameter, were placed over the mid-portion of the pons on 3–4 consecutive slices of the SPGR (Lucignani et al., 2004). Then, using the RPM software package, the SRTM procedure was applied and voxel-by-voxel BPND maps generated (Gunn et al., 1997; Lucignani et al., 2004).


The first step was cross-modality image co-registration. The SPGR scan was co-registered to the acute plain CT, using Visualization Toolkit Kitware-Computational Imaging Science Group (vtk-CISG). As the two-slice CTp scan was done immediately after the plain CT, it was consistently in the same spatial orientation (all recruited patients were co-operative and none showed movement artefact). The matrix transformation file obtained was applied to the FMZ-BPND map (already co-registered to the MR), therefore placing the acute CTp maps, outcome MRI and BPND maps all in register. To allow this, the individual gantry tilt effect was first removed and the BPND maps were smoothed over 9 mm using a box kernel in the z dimension to place the BPND and CTp maps in a similar z resolution. Finally, the BPND maps, CTp maps and T1-SPGR were re-sliced to the original FMZ voxel size in the x–y plane as 2.34 × 2.34 mm.

Following this co-registration step (see Fig. 1 for an illustration of the results), mean transit time (MTT) maps (in seconds) were generated using standard deconvolution (Wintermark et al., 2001). We elected to use the MTT rather than cerebral blood flow (CBF) because although the latter can be determined by CTp with deconvolution analysis (Gillard et al., 2001; Wintermark et al., 2001; Kudo et al., 2003), time-based variables are the most reliable perfusion parameters when using intravascular tracer tracking methods (Wintermark et al., 2005). The slice containing the clearest MCA branches on the non-affected hemisphere was selected, and a voxel with the expected AIF shape was manually selected. A representative arterial curve was then constructed automatically based on those voxels having a value of at least 80% of the chosen voxel in a 1 cm radius across all time points of the selected voxel.

Fig. 1

Illustrative findings from two patients. Each panel shows a representative slice from the co-registered acute plain CT, MTT map, outcome T2- and T1-weighted SPGR and FMZ-BPND map. To be consistent with the SPM analysis in Talairach space, images have been flipped so the left hemisphere is shown on the left of the image, i.e. in the neurological convention. Clinical details can be found in Table 1. Patient 1 showed an extensive MTT deficit in the affected MCA territory on CTp performed 60 min post-onset, but made an excellent recovery following i.v. t-PA with a very small outcome basal ganglia infarct and a normal BPND map. Patient 2 also had an extensive, though more profound, MTT abnormality on CTp done 120 min after stroke onset and made an excellent recovery with a small basal ganglia infarct, but showed definite reductions in BPND in the non-infarcted MCA cortex.

Data analysis

Comparison of BPND and VT maps

For each control, 110 circular ROIs were placed over the cortical mantle across 10 slices on the SPGR, and the mean VT and BPND values for each ROI obtained. Linear correlations using VT as gold-standard were computed for each control.

Infarct ROI

Using MRIcro (www.psychology.nottingham.ac.uk/staff/cr1/mricro.html), Infarct ROIs were delineated on the T2-weighted and SPGR images by an experienced investigator (JCB) blinded to all other data.

Voxel-based analysis of SNL

First, the SPGR data sets were spatially normalized to the SPM2 T1 template, and the transformation parameters applied to the BPND maps. This was carried out using infarct masking (Brett et al., 2001) to reduce incorrect warping. To improve sensitivity, the BPND maps were then proportionally scaled to the mean of the unaffected hemisphere, obtained using the SPM2 thresholded mean voxel value. Prior to SPM the FMZ data were smoothed by a 12 mm Gaussian kernel. The analysis was applied within a MCA mask created using a combination of Damasio and Phan's maps (Damasio, 1983; Phan et al., 2005), with the Infarct ROIs subtracted as described below. The analysis was performed within the final mask using Wake Forest University Pick Atlas (WFU PickAtlas; Maldjian et al., 2003) applying small volume correction. The following analyses were performed:

(i) Single-subject analysis: In order to allow an objective within-subject comparison between the non-infarcted MCA territory on the affected side and the mirror mask on the non-affected hemisphere, each patient was first compared with the 10 controls (contrast: patient < controls), using P < 0.001 uncorrected as cut-off, which was determined based on a prior permutation analysis (Signorini et al., 1999) on the 10 controls that showed no voxel exceeding this statistical cut-off in any subject. This cut-off was then applied to each patient to test the hypothesis that across patients there would be significantly more voxels exceeding this cut-off in the stroke-affected side as compared to the unaffected side. For this analysis, the individual's Infarct ROI was subtracted from the MCA mask so that the analysis would be performed only on non-infarcted areas; prior to doing this, however, the Infarct ROI was dilated by a 14 mm sphere (i.e. twice the effective PET resolution) in order to account for the partial volume effect (PVE), which may smear the infarct data into neighbouring areas (Hoffman et al., 1979).

(ii) Between-group comparison: Here, the group of patients was compared to the group of controls using a two-sample t-test, testing both the controls > patients and patients > controls contrasts. We used a P < 0.05 FDR-corrected cut-off, with only clusters >1 voxel considered significant. For this analysis the Infarct ROIs of all subjects were subtracted from the common MCA mask. Prior to this they were smoothed using MRIcro implementation of Brett's method (Brett et al., 2001) with FWHM of 2 mm and a P threshold of 0.001 adjusted in the z plane. The voxel clusters exceeding the above statistical cut-off were then projected onto a T1-weighted MRI in standard Montreal Neurological Institute (MNI) space (average of 105 normal subjects) for anatomical orientation.

The above analyses were repeated following voxel-based PVE correction (PVEc), using the Alfano-modified Muller-Gartner method as implemented in the PVE-Out software (Quarantelli et al., 2004). This employs the segmented SPGR images from SPM2, taking into account the effective resolution of the PET scanner, and assuming that the CSF provides zero signal and the white matter is homogenous. The latter assumption is necessary for PVEc of grey matter when using this method (Quarantelli et al., 2004). In support of this assumption, in vitro autoradiographic studies on human brain tissue have consistently shown complete lack of cBZR and FMZ-specific binding in the white matter (Young and Kuhar, 1979; d’Argy et al., 1988). This was borne out in the present data set, where BPND in central white matter was not different from zero and had similar variance in both groups of subjects (data not shown).

FMZ-BPND versus acute-stage perfusion deficit

To test the hypothesis that the more severe the hypoperfusion during the occlusion phase the worse the outcome SNL, we determined both BPND values and MTT in the same sets of ROIs sampling the ultimately non-infarcted cortical MCA territory. Patients whose CTp disclosed no hypoperfusion were excluded. Contiguous circular ROIs 14 mm in diameter were placed manually (using ANALYZE 7.0) on the cortical rim of the two co-registered SPGR slices corresponding to the two CTp slices, but avoiding any overlap with the Infarct ROIs. They sampled the MCA cortical territory as well as the anterior and posterior watershed. To further exclude any influence of the final infarct, the ROI bordering the infarct on each side was discarded. The ROIs were flipped to form equivalent mirror ROIs on the unaffected hemisphere, and transferred to the CTp and BPND maps. To improve sensitivity, all values were normalized to the unaffected side. Thus, for each pair of ROIs the affected/unaffected BPND asymmetry ratio (FMZ-AR) and the MTT delay (affected–unaffected, in seconds) were calculated. Simple linear regression analysis between MTT delay and FMZ-AR was then computed for each patient, and the slope [± standard error (SE)], intercept and correlation coefficient (r) obtained. To test our primary hypothesis, we then assessed whether the slope across individual patients was significantly negative (each subject contributing one independent value only). To assess this, the mean of the slope estimates of the individual patients was tested by reference to the pooled SE (root mean squared SE) of the individual slopes.

These analyses were repeated with ROI-based PVEc applied using PET Analyzer (PETAN; Smielewski, 2003), which uses the data from the segmented MR to calculate within each ROI the BPND values for brain, corrected for cerebrospinal fluid (Meltzer et al., 1999).



Seven consecutive patients were studied (three males/four females; age: 43–75 years). Their main demographics and clinical and radiological data are shown in Table 1. Only Patient 1 was known to be hypertensive before the stroke. All presented <6 h from onset with moderate-to-severe left MCA territory stroke (NIHSS range: 8–24), and minimal changes on plain CT, with none having >1/3 MCA territory involved. The CTp scans disclosed an extensive perfusion deficit in 6/7 patients, over the whole MCA territory in 5 (see Fig. 1 for two illustrative examples) and over the posterior MCA territory in one (Patient 6), while in Patient 4 it revealed a large area of abnormally short MTT in the posterior MCA territory, indicating spontaneous recanalization prior to CTp. Five patients received recombinant tissue plasminogen activator (t-PA). All seven patients had good early improvement (ΔNIHSS ≥ 6 at 24 h) and good final recovery (ΔNIHSS on the day of the PET study: 6–23). The structural MRI and PET were obtained on the same day, 20–415 days (median: 36) post-stroke. No patient had significant white matter changes on T2-weighted MRI. The Infarct ROIs projected onto each subject's structural MRI are shown in Fig. 2, and infarct volumes in Table 1: 6/7 patients had small-to-medium size subcortical infarcts (occasionally involving part of the insula), and Patient 6 had a moderate-sized parietal infarct.

Fig. 2

Individual Infarct ROIs (see Material and methods section for details) displayed on the patient's own T1-weighted MRI. To be consistent with SPM analysis in Talairach space, images have been flipped so the left hemisphere is shown on the left of the image, i.e. in the neurological convention.

View this table:
Table 1

Demographics, clinical and radiological data

PatientAge/sexInitial NIHHSPlain CTCTp: minutes since onsetCTp findingst-PANIHSS 24 hDate of PET (days)NIHSS at PET studyInfarct topography (volume, ml)Stroke mechanism
173/M20EIS + HMCAS60Ext. MTT lesionaYes14460L MCA S/C (5.7)Cardioembolic
275/F23EIS + HMCAS120Ext. MTT lesionNo4202L MCA S/C (13.9)Atherothrombotic
359/F18EIS + HMCAS120Ext. MTT lesionYes51820L MCA S/C (11.4)Cardioembolic
443/M13Subtle EISAwakeningShort MTTbNo4333L MCA S/C (3.2)Cryptogenic
573/F13Subtle EIS + HMCAS240Ext. MTT lesionYes6305L MCA S/C (10.4)Cardioembolic
669/M8EIS165Post. MTT lesionYes2362L MCA Parietal (32.2)Cardioembolic
756/F24EIS + HMCAS115Ext. MTT lesionYes74151L MCA S/C (14.5)ICA stenosis >70%
  • aMTT Lesion signifies an area with markedly prolonged MTT indicative of acute ischaemia. bIn this patient there was no MTT lesion, but an area of abnormally short MTT in the posterior L MCA territory, indicative of spontaneous recanalization.

  • EIS = Early ischaemic signs; HMCAS = hyperdense MCA sign; Ext. = extensive; Awakening = awakening stroke so time of CTp not clear; L MCA = left middle cerebral artery; s/c = striatocapsular; ICA = internal carotid artery; Post. = posterior.

FMZ in controls

The VT and BPND maps showed the expected high binding in cortical areas, moderate binding in basal ganglia, thalamus and cerebellum, and lack of binding in the white matter, with similar tracer distribution in both sets of images. Both BPND and VT maps were available in nine controls (one volunteer declined the arterial line). As expected there was an excellent linear correlation between the BPND and VT values in each subject (r range: 0.957–0.999, all P < 0.0001; data not shown).

FMZ results in patients

Two illustrative patients are shown in Fig. 1. Both exhibited an extensive perfusion deficit (more profound in Patient 2) with mild early CT ischaemic signs and a small final infarct, but Patient 1 had essentially normal FMZ scan whereas Patient 2 had a clear-cut reduction in BPND in the affected but non-infarcted MCA cortical territory.

Individual SPM analysis

Figure 3 shows two representative slices from each patient's results. Voxels exceeding the pre-specified statistical cut-off were present in the affected (left) hemisphere in all patients but one (Patient 1), and in the opposite hemisphere in four patients (1, 3, 5 and 6). The sum of significant voxels in the right and left MCA masks (minus dilated Infarct ROIs) are also shown in Fig. 3. In only one patient (1) was this sum larger on the unaffected than the affected hemisphere, and it was small. This distribution was statistically significant (P = 0.028, Wilcoxon sign rank test). Repeating this analysis on PVEc BPND maps, the number of significant voxels was as expected lower in both hemispheres but the overall pattern of distribution was maintained (Wilcoxon P = 0.07).

Fig. 3

Individual SPM results (FMZ-BPND data, not corrected for partial volume effect). Two representative MR slices from each patient are displayed in the neurological convention. The Infarct ROI is displayed in green. Voxels with significantly low FMZ-BPND beyond the pre-specified statistical cut-off (P < 0.001 uncorrected as compared to controls, see Material and methods section) within the MCA mask (minus dilated Infarct ROI, see Material and methods section) are shown in yellow/orange for the affected hemisphere, and blue for the opposite hemisphere. The sum of all significant voxels on the affected and unaffected hemispheres is shown below each patient's images. Significant voxels were consistently more numerous on the affected as compared to unaffected side in all patients but one (1) where they were present on the unaffected side only, but in small number. This distribution is statistically significant (P = 0.028, Wilcoxon signed rank test).

Between-group SPM analysis

Between-group SPM comparisons demonstrated significant and extensive reductions in FMZ-BPND in the patient group relative to the controls (Fig. 4). The clusters were chiefly located in the peri-insular, superior temporal and prefrontal regions. No clusters surviving the pre-specified threshold were found in the opposite hemisphere. Table 2 lists the SPM peaks. There was no finding with the reverse contrast. Repeating the SPM analysis on PVEc BPND maps showed no substantial change, with peak significances not meaningfully reduced (data not shown).

Fig. 4

Between-group comparison, SPM results. Images are displayed in the neurological convention. The clusters (yellow blobs) that exceeded the predetermined statistical cut-off (FDR P < 0.05 corrected) within the MCA mask minus the Infarct ROIs are projected onto a T1-weighted MRI in standard MNI space. Clusters were found exclusively on the affected (left) side, and were located chiefly in the peri-insular, superior temporal and prefrontal regions. See Table 2 for details.

View this table:
Table 2

Results of SPM between-group comparison (controls > patients, P < 0.05 FDR-corrected)

Talairach labelCluster voxel numberVoxel P (FDR-cor)Voxel TZx (mm)y (mm)z (mm)
Superior Temporal Gyrus (BA39)17880.0087.014.60−46−4912
Middle Temporal Gyrus (BA22)0.0085.964.20−46−24−4
Superior Temporal Gyrus (BA22)0.0085.764.12−55−318
Middle Frontal Gyrus (BA10)13710.0086.524.42−424112
Middle Frontal Gyrus (BA6)0.0085.804.14−44448
Superior Frontal Gyrus (BA6)0.0085.403.96−262052
Superior Frontal Gyrus (BA10)590.0114.403.47−225224
Caudate body520.0144.113.31−12316
Middle Frontal Gyrus (BA10)150.0154.013.25−3054+1
Middle Frontal Gyrus (BA10)0.0353.162.72−28538
Caudate Body40.0353.182.74−12−1520
Inferior Frontal Gyrus (BA47)80.0363.142.71−2415−16
Superior Temporal Gyrus (BA38)390.0373.122.70−445−20
Precentral Gyrus (BA44)20.0412.992.61−46128
  • Bold face indicates primary SPM results for each cluster/peak.

Relationship between FMZ-BPND and MTT delay

This analysis was run on the data from the six patients with CTp-documented hypoperfusion (i.e. Patients 1–3 and 5–7, Table 1). In this sample, the affected-side MTT delays ranged from −0.4 to 7.78 s. Individual linear regressions between FMZ-AR and MTT delay showed negative slopes in five patients, statistically significant in four (2 and 5–7), while in the remaining patient (1) the slope was weakly positive but with a large SE and thus non-significant (see Fig. 5 for details). The mean slope and pooled SE was −0.0341 ± 0.0071 (Z = −4.8, P < 0.001). The scatter plot of the individual patients and the mean slope are shown in Fig. 5, which illustrates the overall decline in FMZ-AR with increasing MTT delays across patients, with MMT delays beyond 5.5 s always associated with FMZ-AR values <1. Applying ROI-based PVEc as expected resulted in a marked increase in FMZ-AR data variance yet the overall results were maintained (mean slope = −0.0192 ± 0.012, P = 0.11).

Fig. 5

Scatter plot of the relationship between FMZ asymmetry ratio (FMZ-AR) and MTT delay (as determined for pairs of ROIs) in non-infarcted MCA cortical areas in Patients 1–3 and 5–7 (colour and symbol code is shown in the right-hand side panel; Patient 4 was excluded from this analysis because his CT perfusion scan did not show presence of hypoperfusion in the affected MCA territory), showing the linear regression for individual patients and the mean slope across patients, which was significantly negative (P < 0.001). The equation of the regression lines (y = ax + b), the slope (± SE), intercept, correlation coefficient (r) and P-value for each patient are shown in the right panel. This scatter plot illustrates the overall decline in FMZ-AR with increasing MTT delays across patients, with MMT delays beyond 5.5 s always associated with FMZ-AR values <1.


Although this small-scale ‘proof-of-principle’ study should be considered preliminary, the results show the presence of highly significant and widespread reductions in FMZ binding in the non-infarcted but initially hypoperfused MCA territory, and a significant relationship between acute-stage MTT delay and subsequent FMZ binding, in keeping with earlier animal studies. Finally, these results were maintained after controlling for the potentially confounding effect of tissue loss.

Mirroring the penumbra concept, FMZ binding was assessed as a surrogate of SNL in the subacute-to-chronic stage (see Introduction section). Based on previous evidence (Nakagawara et al., 1997), patients were scanned from 3 weeks onwards. However, one may argue that the observed loss of FMZ binding was not the direct consequence of the acute event, but occurred later from other causes. Several lines of evidence speak against this contingency: (i) the occurrence of any TIA or stroke after the index stroke was an exclusion criterion; (ii) most patients had cardioembolic strokes and subsequently received appropriate treatment and (iii) the observed correlation between MTT delay and FMZ-binding points strongly to a direct relationship. Note that in both previous similar clinical studies (Nakagawara et al., 1997; Saur et al., 2006), patients were also scanned in the subacute-to-chronic stage. Although it would in principle be preferable that all scans are done at the same post-stroke delay, achieving this would be virtually impossible due to patient's preferences and availability of the complex technical chain behind quantitative PET. Nonetheless, this variability and lateness in scanning times is a limitation of this study.

Patient 7 was studied the latest and her stroke was ascribed to ipsilateral >70% carotid stenosis. Since symptomatic carotid stenosis has been associated with reductions in ipsilateral cortical FMZ binding (Yamauchi et al., 2005), the PET findings in this patient may be due to the carotid stenosis rather than to her stroke. However, in the above article the FMZ-binding reduction was restricted to the patient group showing cortical border zone infarction, which was absent in this patient. She underwent carotid endarterectomy 2 months after her stroke, and deferred the PET study until long after the operation. However, she did not have any TIA or recurrent stroke after her index stroke. Any haemodynamic effect of the stenosis on perfusion would add on to that from the acute event and would be taken into account in the analysis. Excluding this patient from the analyses was therefore unjustified. At any rate, doing so would not change the overall results (Wilcoxon test on significant voxels: P = 0.046; group SPM analysis: still many clusters at FDR P < 0.05; and mean MTT delay–FMZ ratio slope: 0.0309 ± 0.0082, P < 0.001).

Decreased 123I-IMZ uptake in the cortex overlying large striatocapsular infarcts has been reported in several SPECT studies (Hatazawa et al., 1995; Dong et al., 1997; Nakagawara et al., 1997; Sasaki et al., 1997), although not all patients were affected (Sasaki et al., 1997; Takahashi et al., 1997). Consistent findings have been reported in baboons using FMZ-PET (Sette et al., 1993; Giffard et al., 2008). However, prior to the present investigation, only two clinical studies embarked on the demanding longitudinal assessment of SNL in the ultimately non-infarcted but acutely at-risk tissue (Nakagawara et al., 1997; Saur et al., 2006). The former authors reported reductions of 123I-IMZ binding (mean AR 0.89) 5 days to 23 months post-stroke in cortical areas acutely exhibiting hypoperfusion (or hyperperfusion as a marker of prior ischaemia) on 133Xe-perfusion SPECT but intact on follow-up structural imaging. The latter authors looked at 123I-IMZ uptake in the MR diffusion-perfusion mismatch (with TTP >4 s) that did not progress to infarction; 5–15 days after stroke; the IMZ-AR for the whole non-infarcted mismatch was 0.95 (range 0.89–1.01). In the baboon, Giffard et al. (2008) reported a mean FMZ-AR of 0.90 in the most severely affected ROIs, 21–47 days after temporary MCA occlusion. The range of FMZ-AR found in the present study (Fig. 5) is entirely consistent with the above reports. However, more severe reductions, not assessable here, may occur beyond PET resolution and/or nearer the infarct.

Both Nakagawara et al. (1997) and Saur et al. (2006) however acknowledge the flow dependence and difficult modelling of 123I-IMZ as well as the poor spatial resolution and inadequate quantification of SPECT. Additionally, in the latter study (Saur et al., 2006) the patients were scanned at a time when infarct delineation can be inaccurate due to perifocal oedema, and blood–brain barrier leakage may complicate interpretation due to possible brain passage of labelled metabolites. Most Nakagawara et al. patients were scanned in the chronic stage but volume loss from collapse of infarct cavity was not corrected for, and cross-modality image co-registration was not performed. In contrast, here we quantified cBZR binding ≥20 days after stroke onset using PET and a flow-independent tracer, and implemented full co-registration and PVEc.

In this investigation, we used CTp-based MTT delay to assess perfusion. Studies comparing CTp with deconvolution to PET (Kudo et al., 2003) or Xenon CT (Wintermark et al., 2001) document that CTp allows quantification of CBF with good reproducibility (Gillard et al., 2001). However, as with all intravascular bolus contrast methods, reliability is best for time-based variables such as MTT, particularly MTT delay relative to the unaffected side (Wintermark et al., 2005). Since there is a well-established relationship between MTT and CBF (Powers, 1991), MTT delay was used here as a surrogate for CBF. Using this approach, this study is the first to document a negative relationship between initial hypoperfusion and subsequent neuronal damage, as estimated using FMZ. Nakagawara et al. (1997) did not report on this relationship, while Saur et al. (2006) reported a non-significant trend only, though in the expected direction. This latter inconsistency probably reflects this study's above limitations and also that their analysis involved whole mismatch areas. Our observation was expected from early animal investigations (Jones et al., 1981; Garcia et al., 1983; Heiss and Rosner, 1983; DeGirolami et al., 1984) as well as from a baboon study where areas with chronic-stage FMZ decreases were shown to acutely exhibit hypoperfusion with high oxygen extraction fraction, i.e. ischaemia (Giffard et al., 2008). However, over and above severity of ischaemia, its duration also determines selective neuronal death (Garcia et al., 1997; 1983; Heiss and Rosner, 1983; DeGirolami et al., 1984; Li et al., 2000) but this information was not available in this study. Interestingly, however, Patient 1 who had the least reduced FMZ binding, had his stroke on the hospital grounds and was thrombolysed the earliest (90 min after onset), providing some support to this idea. Future work should aim to determine precisely the time of effective reperfusion.

The relationship observed between putative SNL and severity of acute-stage hypoperfusion is consistent with the general penumbra model, according to which damage is a function of CBF (Astrup et al., 1981; Jones et al., 1981; Heiss and Rosner, 1983). This study suggest the penumbra model may extend to SNL, i.e. penumbral areas that do not progress to frank infarction may develop selective neuronal damage—as if ‘caught’ on the way to pan-necrosis—in proportion to the severity of hypoperfusion (see Fig. 6 for an illustration). Further studies are required to confirm or refute this model. If this relationship is confirmed, it would support the notion that SNL is secondary to ischaemia per se rather than ‘reperfusion injury’, although both may be related and intervention studies would be required to resolve this issue.

Fig. 6

Idealized representation of the classical penumbra concept, here including selective neuronal loss (modified from Jones et al., 1981).

The histopathological counterpart of reduced cBZR binding in the rescued penumbra is still elusive. In the rat, reduced binding of 3H-FMZ (Katchanov et al., 2003) and 123I-IMZ (Watanabe et al., 2000) co-localized with SNL, and in a cat study Heiss et al. (1998) briefly mention mildly reduced in vivo FMZ binding in peri-infarct areas apparently matching a degree of SNL at post-mortem. On the other hand, applying a systematic analysis in baboons, Giffard et al. (2008) found only mild and inconstant histopathological counterparts, but raised potential confounds, notably the difficulties in detecting SNL in the chronic stage in perfused-fixed gyrencephalic brains. In the rat, dead neurones, conspicuous in the first few days, are gradually phagocytosed and replaced by ‘ghosts’ (Nedergaard, 1987; Garcia et al., 1997; Ito et al., 2007). These factors, on top of potentially biased samples, may explain why SNL is only rarely observed in man post-mortem (Lassen et al., 1983; Nedergaard et al., 1984, 1986; Torvik and Svindland, 1986). Furthermore, ultrastructural damage short of actual cell death, such as dendritic spine loss (which may bear the cBZR), is reported following brief ischaemia (Tomimoto and Yanagihara, 1992; Li et al., 2000; Ringer et al., 2001; Zhang et al., 2005). The mild reductions in cBZR binding observed here in the rescued penumbra thus likely reflect a variable combination of SNL proper and subtler cell damage, depending focally on the severity and duration of ischaemia.

Previously, FMZ-PET has been used in the hyperacute stage as a predictor of gross infarction (Heiss et al., 1997, 2001). It is feasible that the nature of FMZ binding loss in the acute ischaemic core may differ from that seen chronically with SNL. For instance, the former may reflect acute disturbances in ionic milieu receptor internalization or release of GABA-blocking ligand binding, while the latter may represent delayed synapse loss from the acute ischaemic insult and/or reperfusion injury.

Our findings have relevance to the re-appearance of DWI lesions following spontaneous or therapeutic reperfusion (Fiehler et al., 2002; Kidwell et al., 2002). In animal models, SNL affects the neocortex as well as the striatum (Garcia et al., 1996; Li et al., 2000), but FMZ is best suited to study the former (Pappata et al., 1988), and cortical damage is most relevant to clinical outcome. In the present study, decreased cortical FMZ binding mainly affected peri-insular, superior temporal and prefrontal regions (Fig. 3), i.e. all areas likely to be penumbral in proximal MCA occlusion (Weiller et al., 1993; Darby et al., 1999; Markus et al., 2003; Phan et al., 2005). Post-stroke subtle cognitive or affective impairments may be partly due to SNL, and peri-infarct SNL may slow down recovery and increase long-term risk of vascular cognitive impairment and mixed dementia (Gold et al., 2007). Thus, therapies to prevent or limit SNL may have short- as well as long-term benefits.

Finally, that SNL can affect the rescued penumbra does not invalidate the penumbra concept nor call into question mainstream recanalization therapy. Even despite potential occurrence of SNL, rescuing the penumbra from complete infarction is the main determinant of clinical recovery (Furlan et al., 1996). Furthermore, as all the patients in this study had good-to-excellent clinical outcome, there is no evidence at this stage that recovery is impeded by SNL. Specific investigations to address this issue are required.


Work supported by Medical Research Council UK (grant number G0001219 and G0500874 to J-CB). The authors thank Prof JD Pickard, Prof DK Menon, Dr JH Gillard, Mrs O. Golovko and Dr M. Takasawa for their support, and Mrs M. Rawlings for help with preparation of the manuscript. P.S.J. and EAW are supported by NIHR Biomedical Research Centre Grant.


  • Abbreviations:
    arterial input function
    central benzodiazepine receptor
    computerized tomography
    CT perfusion
    diffusion-weighted imaging
    fluid-attenuation inversion-recovery
    middle cerebral artery
    mean transit time
    positron emission tomography
    regions of interest
    selective neuronal loss
    single-photon emission tomography
    spoiled-gradient recalled
    simplified reference tissue model
    transient ischaemic attack


View Abstract