Mapping neuronal density in peri‐infarct cortex with PET

Despite modern acute reperfusion therapies aiming to salvage the acutely ischemic but still viable tissue, so‐called “penumbra” [Astrup et al., 1981], stroke remains a major killer and the first cause of adult‐acquired handicap worldwide [Feigin et al., 2016]. Even in thrombectomized patients in whom the occluding thrombus was removed within 6–8 h of stroke onset, less than half recovered to an independent life [Goyal et al., 2016]. Novel targets to therapy as adjuncts to these highly effective but insufficient therapies are therefore needed [Fisher and Saver, 2015]. Mapping the areas of selective neuronal loss (SNL) beyond the actual infarct has attracted considerable interest lately, as preventing SNL could save precious brain resources, facilitate plastic processes including in peri‐infarct areas [Nudo, 2006], and in turn improve functional outcome [Baron et al., 2014].

In their recently published article, Funck et al. [2017) for the first time mapped neuronal density in the immediate (3–15 mm) peri‐infarct cortex (PIC) in 11 stroke patients by means of flumazenil (FMZ) PET, according to a longitudinal design involving scanning both 10–31 days poststroke and again ∼6 months later. To address this challenging issue, they used a sophisticated image postprocessing methodology implementing both geodesic distance metric and local partial volume effects (PVE) correction (PVC) [Funck et al., 2014]. This allowed them to simultaneously “flatten” the folded peri‐infarct sulci and correct for potential poststroke cortical atrophy.

Flumazenil is a benzodiazepine ligand that binds in vivo to the GABA_A receptors—which are widely and densely distributed in the cerebral cortex—and has been used for over two decades to map ischemic neuronal death and infarction in vivo [Sette et al., 1993]. FMZ has also recently been formally validated as a sensitive in vivo marker of SNL against postmortem immunohistochemistry in a rodent stroke model [Ejaz et al., 2013].

Using the above sophisticated techniques, Funck et al. [2017] found reduced GABA_A‐receptor density and normal cortical thickness in the PIC in the early subacute phase, consistent with SNL, and reduced GABA_A‐receptor density together with reduced cortical thickness—i.e., cortical atrophy—in the PIC in the chronic phase. PVC had a significant effect on estimates of GABA_A‐receptor density in the PIC in the chronic stage only, indicating that thanks to PVC, SNL was not overestimated at that stage. Although overall the PVC‐corrected loss in GABA_A‐receptor density in the PIC was greater and tended to expand to longer distances from the infarct borders in the chronic stage, there was no significant effect for any prespecified distance, indicating overall an essentially stable SNL over time. This observation is consistent with our earlier baboon studies showing stable FMZ binding loss in peri‐infarct areas from day 3 onward [Sette et al., 1993], and with human studies by Nakagawara et al. [1997] showing apparent stability of FMZ loss in noninfarcted areas from 2 to 3 weeks onward. Similarly, in Guadagno et al.'s [2008] FMZ study, there was no significant effect of time since stroke onset on FMZ interhemispheric difference in noninfarcted cortex in 7 patients studied 20–415 days poststroke.

In rodent stroke models of temporary ischemia >10–15 mins, including studies from my laboratory [Ejaz et al., 2015a, 2013, 2015b; Hughes et al., 2010], postmortem assessments have documented that the rescued penumbra is consistently affected by widespread and patchy neocortical SNL, as well as linear peri‐infarct SNL (see Baron et al. [2014] for a review). Unlike in vivo studies, postmortem studies are not affected by the problem of PVE from cortical atrophy. In addition, the severity of SNL was consistently found to be related to local perfusion during arterial occlusion [Ejaz et al., 2013; Hughes et al., 2010], suggesting SNL is a function of the degree acute ischemia although expresses itself after reperfusion.

Prior to the recent report by Funck et al. [2017], two clinical FMZ PET studies had suggested the presence of patchy areas of SNL throughout the surviving ischemic penumbra [Carrera et al., 2013; Guadagno et al., 2008]. To test whether SNL affects the ultimately noninfarcted penumbra, these authors prospectively studied patients with penumbra documented on hyperacute perfusion imaging but who enjoyed major neurological recovery reflecting early reperfusion. The patchy pattern of SNL was thought to reflect the fact that, rather than adopting the classic concentric pattern [Astrup et al., 1981; Baron, 1999], the penumbra often follows a patchy pattern across the affected territory [Furlan et al., 1996; Ogata et al., 2011], probably as a result of local variations in leptomeningeal collaterals density.

Although it has always been assumed from animal and human postmortem studies that SNL would be particularly marked in the immediate PIC (see Baron et al. [2014] for review), neither study attempted to assess SNL in this region because of the expected major PVE from the low FMZ binding in the infarct itself. Thus, in both studies, the infarct mask was “dilated” by around twice the spatial resolution of the PET images, and this mask was excluded from data analysis. Nevertheless, in Guadagno et al.'s [2008] study, PVC based on segmented coregistered MRI was carried out in circular ROIs spanning the entire MCA territory (save for the dilated infarct mask). Although as expected statistical significance was reduced as compared to non‐PVC data, corrected FMZ binding was reduced in many ROIs in most patients and significantly correlated with acute perfusion deficit severity, similar to the above‐cited rodent findings [Ejaz et al., 2013; Hughes et al., 2010]. However, this PVC method corrected only for spill‐out effects, namely, artefactually reduced FMZ binding due to enlarged CSF spaces—i.e., cortical atrophy—but not from spill‐in effects, that is, artefactually increased FMZ binding due to neighboring areas with high binding. The PVC method developed by Funck et al. [2014] affords correction from both effects simultaneously, and therefore provides even more accurate estimates of cortical SNL. However, in Guadagno et al.'s study, individual subject PVC voxel‐based statistical analysis was also applied using the Alfano‐modified Muller‐Gartner method [Quarantelli et al., 2004], which confirmed the presence of patchy SNL in noninfarcted cortical areas. Finally, in neither study was there evidence on inspection of MR images of cortical atrophy in areas exhibiting reduced FMZ binding [Carrera et al., 2013; Guadagno et al., 2008]. Overall, therefore, there is no reason to suspect that the reduced FMZ binding found in the surviving penumbra in these two studies could merely reflect uncorrected PVE from cortical atrophy.

The mechanism(s) underlying the development of cortical atrophy after stroke has lately been the matter of intense research. Although cortical atrophy might directly result from SNL, this has not been directly demonstrated as yet. For instance, in our rat studies that involved pure SNL without actual infarcts, no evidence of cortical atrophy was observed even up to 60 days poststroke—a very long time interval for rodents [Ejaz et al., 2015a]. This likely reflects the fact that SNL was present mainly in patches rather than across extensive areas, and that within these patches the extracellular matrix was intact, indicating preserved tissue architecture [Ejaz et al., 2015a]. In addition, proliferation of microglia (initially) and astrocytes (chronically) could compensate for any SNL‐related tissue volume loss. Over and above SNL, the development of cortical atrophy after stroke likely also reflects delayed Wallerian and (trans)synaptic degeneration secondary to stroke‐induced fiber severage and neuronal death, which can be quite widespread in rodent stroke models [Baron et al., 2014; Kataoka et al., 1991; Zhang et al., 2012] and subcortical stroke patients [Duering et al., 2012; Jouvent et al., 2008; Preul et al., 2005], and would be expected to predominate in the PIC due to direct cortico‐cortical deafferentation. A further, compounding effect might occur after stroke, namely geometric distortion due to brain collapse following infarct cavitation and shrinkage. Importantly, both secondary degeneration and brain collapse would be expected to largely depend on infarct size, and in this line we note that we studied almost exclusively small infarcts [Carrera et al., 2013; Guadagno et al., 2008], whereas several patients in the Funck et al.'s [2017] study had large infarcts (their fig. 1), and accordingly their assessment of SNL might have been affected by both effects.

Whether SNL impacts functional outcome in man is still uncertain. However, in rodents, pure cortical SNL does affect sensori‐motor behavior [Ejaz et al., 2015a; Sicard et al., 2006], and severe striatal SNL affects sensori‐motor [Balkaya et al., 2013], cognitive [Winter et al., 2004], and affective [Winter et al., 2005] behavior. Only indirect evidence for this exists in man so far. For instance, in our study combining acute‐stage perfusion mapping and subacute‐stage FMZ PET and fMRI in patients with small infarcts but good overall recovery, we found that cortical SNL remote from the infarct significantly impaired neuronal activation [Carrera et al., 2013].

In sum, previous work has consistently documented the presence of SNL throughout the salvaged penumbral cortex in both rodents and man. Funck et al. [2017] now document that SNL also as expected affects the immediate PIC, and they should be commended for taking up the major technical challenge involved. Together, the available literature establishes the presence of widespread cortical SNL after ischemic stroke in man, which has important implications for interventions aiming to improve functional outcome over and above the established benefits from early recanalization.