Sepsis and septic shock are associated with hyperglycaemia and peripheral insulin resistance 1 2 . Glycemic control strategies are commonly instituted as adjunctive therapeutic measures in critically ill patients, although recent studies have not consistently shown a benefit from intensive insulin therapy 3 4 5 6 . One argument supporting blood glucose control is that intensive insulin therapy is associated with a protective effect on the peripheral and central nervous system 7 . While it has been shown that intensive insulin therapy reduces the incidence of critical illness neuromyopathy 7 8 and that hyperglycaemia worsens brain injury in ischemic stroke 9 10 11 and head trauma 12 , the effect of hyperglycaemia or insulin on sepsis related brain dysfunction is not well understood. A recent in vitro study showed that hyperglycaemia increased microglial vulnerability to lipopolysaccharide (LPS) mediated toxicity 13 , through formation of oxidative free radicals. Interestingly, it has also been shown that experimental sepsis induces oxidative damages in the brain 14 . In a previous neuropathological study, we found that septic shock is associated with neuronal ischaemia, microglial activation and apoptosis as well as neuronal apoptosis, which was statistically correlated with endothelial expression of iNOS 15 . However, the relationships between BG and neuropathological findings have not been thoroughly assessed. The objective of the present study was to address this issue and also to assess whether hyperglycaemia is associated with neuronal or microglial apoptosis after adjustment to other pro-apoptotic factors. We also evaluated brain expression of Glucose Transporter (GLUT) proteins given their role in transmembrane glucose transport in neurons and microglial cells during stress conditions 16 . Assessment of the relationships between hyperglycaemia and neuropathological abnormalities might provide insight on the mechanisms of sepsis associated neurological and psycho-cognitive long-term consequences.

From 1997 to 2001, 17 patients who died from septic shock were included. Patient characteristics are presented in Table 1. Septic shock had a median duration of four days and was mainly secondary to pneumonia or cellulitis. Four patients had pre-existing diabetes mellitus. Median BG over ICU stay was 2.17 g/l. Episodes of hyperglycaemia were observed in all patients and hypoglycaemia occurred in five (29%) patients. Nine (53%) patients developed prolonged hyperglycaemia and six (35%) were treated with insulin (with mean BG level of 2.7 g/L (1.9 to 3.0)). Macroscopic findings were ischaemia (n = 12), haemorrhage (n = 9) and disseminated abcesses (n = 3). Oedema was observed in only one patient.

In contrast to HLA-DR, expression of microglial CD68 tended to be correlated with AUBGC > 2 g/l (rho = 0.44, P = 0.08). Intensity of neuronal and microglial apoptosis was correlated with AUBGC > 2 g/l (rho = 0.53; P = 0.03 and rho = 0.70; P = 0.002) (Table 2, Figure 1). Intensity of neuronal beta-APP expression correlated with AUBGC > 2 g/l (rho = 0.61; P = 0.03) (Figure 2). Endothelial iNOS expression was correlated with intensity of neuronal apoptosis (rho = 0.68, P = 0.005) but not with that of microglial apoptosis (rho = 0.34, P = 0.17). The intensities of neuronal and microglial apoptosis were correlated (rho = 0.56, P = 0.02). Immunostaining of GLUT3 was not satisfactory. GLUT1 rather stained endothelial cells than neurons and its expression did not vary among patients. Neuronal GLUT4 (Figure 3) and microglial GLUT5 expression (Figure 4) did not correlate with prolonged hyperglycaemia nor with neuronal or microglial apoptosis (Table 3). Expressions of endothelial iNOS and microglial GLUT5 were inversely correlated (rho = -0.54; P = 0.03). Neuronal and microglial apoptosis were not correlated with SAPS-II at admission, highest SOFA score, duration of septic shock, or with serum sodium (especially hyponatremia), lowest systolic arterial pressure, PaO₂ and SaO₂. Intensity of neuronal apoptosis and ischaemia tended to be correlated with cumulative time of hypotension (rho = 0.45, P = 0.06 and rho = 0.38, P = 0.11).

In patients dying of septic shock, hyperglycaemia was associated with microglial apoptosis while neuronal apoptosis was preferentially associated with endothelial iNOS expression. We also found that hyperglycaemia tended to be correlated with CD68 expression, which is a marker of microglial activation. The postulated relationship between hyperglycaemia and microglial cell apoptosis was supported by its absence of statistical correlation with hypotension, hypoxemia or hypernatremia, while it is known that hippocampus is highly vulnerable to these factors. We also found that neuronal GLUT4 and microglial GLUT5 expressions were not correlated with blood glucose level, suggesting impaired downregulation.

These results are consistent with several experimental studies. Discrepancy between microglial CD68 and HLA-DR immunostaining has been previously observed 25 and was ascribed to the fact that CD68 is a better marker of activated microglia. Nitric oxide has been extensively documented as pro-apoptotic factor, notably in experimental sepsis 26 27 28 . In experimental models of cerebral trauma or ischaemia, hyperglycaemia has been linked to neuronal and glial cell injury through various mechanisms including mitochondrial dysfunction, oxidative stress, inflammation and excitotoxicity 29 . Although the similar mechanisms have been implicated in sepsis associated encephalopathy, the potential contribution of hyperglycaemia had not been elucidated. It was recently shown that high glucose and LPS synergistically induce microglial apoptosis by enhancing formation of oxidative free radicals 13 . Interestingly, the statistical correlation between neuronal and microglial apoptosis suggest that they are interdependent phenomenon. It is established that neuronal function and survival is intimately linked to both astroglial and microglial cells 30 . Therefore, one may speculate that hyperglycaemia induces microglial death that, synergistically with endothelial iNOS, induces neuronal apoptosis, suggesting a mechanistic sequence to account for sepsis associated brain dysfunction. This model takes into account the inflammatory 23 and metabolic (hyperglycaemia) pathways that are a major pathophysiological process and disturbance of septic shock, respectively. The correlation between hyperglycaemia and axonal beta-APP expression is consistent with that reported in experimental brain ischaemia 31 . It suggests also another scenario in which hyperglycaemia would first induce axonal injury, then secondary degeneration of microglia 31 . Interestingly, this finding proposes a new pathophysiological mechanism for the long-term cognitive decline in septic patients 32 .

The present study is the first to describe the neuropathological consequences of hyperglycaemia in patients who had died from septic shock. However, our study has several limitations. First, one may argue that apoptosis was rather a post-mortem phenomenon. Although this possibility cannot be ruled out, we have previously shown that cell death did not correlate with time to brain sampling 15 . Second, since BG levels were not assessed continuously, it is likely that discrete hypoglycaemic or hyperglycaemic events were not detected. However, the rate of BG assessment was not different between patients with and without hyperglycaemia or prolonged hyperglycaemia. Third, it has been shown that the capillary test does not provide an accurate measurement of BG, notably overestimating it 33 . However, despite this flaw, capillary meter is used both in clinical trials and in routine for titrating insulin therapy. It has to be noted that microglial apoptosis was also correlated with median BG. Fourth, we have limited our investigation to the hippocampus as it is highly sensitive to hemodynamic, hypoxic or metabolic insults but also involved in ICU associated delirium pathophysiology 34 35 . The impact of neuronal and microglial apoptosis on hippocampal function cannot be obviously inferred from these simple neuropathological observations. It would be of interest to determine experimentally if hyperglycaemia is associated with alterations in hippocampal electrophysiological function and with cognitive impairments mediated by hippocampal structures. It has been reported that high glucose level is associated with occurrence of delirium in ICU patients 36 . Conversely, it has been shown that infusion of glucose is a memory enhancer in septic rats, suggesting that glucose tight control, or at least hypoglicaemia, may affect hippocampal functions 37 .

While we have demonstrated an association between hyperglycaemia and cell death in the brains of septic shock patients, these data do not allow us to make any definitive conclusions on hyperglycaemia as a causative mechanism for cell death. Indeed, statistical correlations between ante-mortem variables and post-mortem findings do not prove a causal relationship. Demonstration of such a link would require a more detailed investigation of how glucose levels affect microglial cellular and molecular function and the demonstration that glycemic control reduces microglial apoptosis. Only experimental studies could reasonably address these issues. Indeed, post-mortem examination does not yield insight into the proximal processes that precede apoptosis in humans. This may explain that hyperglycaemia tended to be correlated with microglial activation (reflected by CD68 expression), which is prior to apoptosis. Assessment of neuropathological effect of BG control would require brain sampling in patients who had died from septic shock and who had or not been treated with insulin therapy: a task not so easily achievable. Our neuropathological samples were obtained before the widespread implementation of glycemic control with intensive insulin therapy in many critical care units. This is illustrated by the fact that insulin was administered in a small proportion of patients and was not targeted to normoglycaemia. These observations prevented us from assessing the neuropathological effect of insulin. Moreover, anticipating a neurological benefit from insulin therapy is premature. First of all, even if microglial cells play a major role in host defence of the brain, and are involved in neuroinflammatory and neurodegenerative processes, their implication in sepsis related brain dysfunction is not demonstrated 38 . It is unknown whether microglial apoptosis is an adaptive, negligible or deleterious phenomenon. Unlike the situation in neurons, interpretation of positive ISEL staining in glial and microglial cells is not straightforward. As ISEL is not absolutely specific for double-stranded DNA breaks and can also detect single-stranded breaks as observed in cell multiplication 39 , positive staining may also reflect cell proliferation. On the other hand, Petito and Roberts 40 suggested that apoptotic death of reactive astrocytes might be a physiological mechanism whereby the brain removes an excess number of astrocytes that have proliferated after certain types of brain injury. This can also apply for microglia 41 . Second, cerebral glucose metabolism is highly complex and its disturbances in sepsis insufficiently elucidated. Therefore, neuronal sensitivity to hypoglycaemia and hyperglycaemia might be deeply changed in sepsis, making the effect of insulin on neuronal metabolism unpredictable. We have found that neuronal GLUT4 and microglial GLUT5 expression were neither correlated with blood glucose levels or cell apoptosis. This does not rule out that glucose transporters are involved in cell death process. For instance, it has been experimentally shown that GLUT5 is implicated in hyperglycaemia-related microglial cell death 13 . Furthermore, one may have expected that glucose transporter expression would have been inversely proportionate to blood glucose level 40 42 43 . Therefore, it is conceivable that its absence of downregulation might have increased intracellular glucose concentration and, thereby, its toxicity. We acknowledge that absence of correlation between microglial apoptosis and GLUT expression does not rule out an alteration of GLUT functioning, which in future studies could be indirectly evaluated by measuring intracellular glucose load and protein glycation. Additionally, it is biologically plausible that hypoglycaemia potentially is far more harmful for the brain than hyperglycaemia. It will be worthwhile to assess the neuropathological correlates of hypoglycaemia in patients who had died from septic shock. This would require a greater proportion of patients who had developed hypoglycaemia than that observed in the present study. It is interesting to note that iNOS has been shown to decrease cerebral GLUT1 expression 44 . One may argue that the slight GLUT1 immunostaining of neurons reflects a downregulation. Although expression of GLUT3 could not have been assessed for technical reasons, it has to be noted that alteration of GLUT3 cannot account for the relationship between hyperglycaemia and apoptosis microglial cells as it is not expressed by these cells.

The present study suggests a similar effect on microglial GLUT5 expression. Other mechanisms could be involved, especially perivascular edema that can compromise substrate and oxygen delivery. Although we have not specifically assessed this mechanism, it is established that the BBB is altered in experimental sepsis but also in septic shock patients 45 .

Despite these limitations, our study suggests that hyperglycaemia may contribute to the complex web of abnormal signalling, which causes sepsis associated brain dysfunction. Future studies should investigate the mechanisms of hyperglycaemia related microglial apoptosis, particularly the impaired downregulation of GLUT, and assess the neuropathological as well as neurological effects of BG control by insulin therapy.

It appears likely that hemodynamic, inflammatory and metabolic factors contribute to brain cell dysfunction and death during septic shock, and may account for sepsis associated brain dysfunction, which is associated with increased mortality 46 . More research is needed to understand the pathogenic significance of these factors and how they may be modulated to therapeutic ends.

• In septic shock patients microglia is strongly activated.

• Hemodynamic, inflammatory and metabolic factors contribute to brain cell dysfunction and death during septic shock.

• Hyperglycaemia is associated with microglial apoptosis while neuronal apoptosis is preferentially associated with endothelial iNOS expression.

• Hyperglycaemia may contribute to the complex web of abnormal signaling which causes sepsis associated brain dysfunction.

iNOS: inducible Nitric Oxide Synthase; AUBGC: area under the BG curve; Beta-APP: amyloid precursor protein A4; BG: blood glucose; GLUT: glucose transporter; IQR: interquartile range; ISEL: in-situ end labelling; LPS: lipopolysaccharide; SAPS II: Simplified Acute Physiologic Score II; SOFA: Sequential Organ Failure Assessment; TNF-α: tumor necrosis factor α.

The authors declare that they have no competing interests.

AP conceived the study, acquired data and wrote the manuscript. JPB helped in interpretation of the data and in drafting the manuscript. RP participated in the design of the study, performed the statistical analysis and helped to draft the manuscript. RS helped to draft the manuscript. SS helped in acquisition of data and revising the manuscript, while RDS also helped to revise the manuscript. CG, FC, FG, DA and VM helped to draft the manuscript. GLG helped in acquisition and interpretation of data. TS conceived the study, participated in the design of the study and helped to draft the manuscript. All the authors read and approved the final manuscript.