The main finding of our study was a global GM volume decrease during participation in an ultramarathon. This change was paralleled by a decreasing body weight. Both changes reversed to the baseline when measured about 8 months after the end of the race. DWI and FLAIR imaging revealed no new brain lesions during the race.
Observed loss in GM volume of about 6% during the 2 months of the race would equal (assuming linearity) an annual rate of 36%, and therefore appears substantial when compared with annual rates of volume losses associated with natural processes: aging leads to volume reductions of GM of less than 0.2% annually [18, 19], Alzheimer's disease shows up to 2% loss per year , and multiple sclerosis leads to annual atrophy rates of about 0.5% . At present, we can only speculate about the reasons that may be responsible for the observed loss in GM volume although the search for possible factors is constrained by two imminent characteristics: a substantial reduction during ultramarathon activity that was reversible upon follow-up. Therefore, possible factors should be consistent with this observation, especially reversibility of volume reduction.
Among the various factors causing brain volume changes , some major causes such as intake of alcohol or toxic substances can certainly be ruled out. Other factors such as inflammation and edema, changes in electrolyte balance, vascular permeability and dehydration, as well as protein catabolism cannot easily be discarded, and also systemic illnesses and corticosteroids must be taken into account . For example, brain volume loss has been shown in illnesses such as kwashiorkor , Cushing's syndrome , and anorexia nervosa .
Body fat reduction due to the huge energy deficit incurred over the course of the race  has been shown in different multistage endurance events [34, 35, 47]. Besides fat loss, in one report on a multistage ultramarathon over 1,200 km a reduction of muscle mass was also noted . Therefore, catabolism with reduction of fat and muscle mass has to be expected during a multistage ultramarathon and may represent a relevant factor for GM volume loss which is strongly supported by the apparently parallel loss of body weight (see Figure 4).
Under physiological conditions, the brain controls the calorie intake to secure a steady supply of necessary nutrients . However, during phases of catabolism protein loss and hereby reduction of colloidal osmotic pressure and a shift of fluid to the subarachnoid spaces  can lead to a shrinkage of the whole brain, which may have contributed to a decrease of GM volume.
Recently it also has been suggested  that elevated cortisol levels might be responsible for GM reduction in patients with anorexia nervosa. The volume loss was seemingly reversible with reported recovery  after successful treatment. This has also been shown for hypercortisolism-induced brain atrophy  and for hippocampal atrophy that appeared reversible after hormone normalization . Interestingly, some recent studies could show that endurance sports increase cortisol levels [50, 51]. Hence, among those various conditions associated with GM volume reduction, hypercortisolism appears to be a likely candidate given that daily running with average distances of about 1.7 marathons activates the hypothalamic-pituitary-adrenal (HPA) axis.
Although dehydration has been shown to relate to a brain volume reduction of about 0.55%, which was reversible upon rehydration , the extent of these changes does not explain the changes of about 6% in our present study. Furthermore, our athletes were extremely well trained and were sufficiently provided with fluid throughout the entire race. Regardless, hyponatremia has been shown in marathon runners and has been linked to hypotonic encephalopathy  or brain edema . However, hyponatremia among multistage ultramarathoners is rare according to a recent report , and our analysis of diffusion and T2-weighted MRI images did not show any new lesions or signs of edema in our sample. In presence of an excellent inter-rater reliability we detected only pre-race lesions in FLAIR imaging, representing older lesions (glial scars), which are expected to show up on each ensuing examination. Even though diminishing visibility of edema and glial scars during a phase of hypercortisolism might seem plausible, the numerical decrease of the average number of lesions on FLAIR imaging in our raw data was an artifact due to varying attendance of subjects (see Table 3 and Additional file 1, Tables S1-3 for paired t test comparisons). Therefore no statistically significant variations in the number of lesions were observed.
This absence of the formation of new brain edema also supports that hypoxic disruption of the brain barrier , or arterial hypertension (which has been linked to reversible posterior brain edema [10–12] in previous studies) are rather unlikely to have occurred during the race in our subjects included. Similarly, disturbances of intravascular coagulation known to happen in marathon runners [13, 14] may produce focal lesions, but no new lesions were detected during the TEFR09 race.
What really distinguishes participants of TEFR09 from leisure athletes normally participating in marathon events is the amount of training they undertake: the TEFR09 participants had run a mean of 5,523 km (SD 1,874 km, range 2,500 km to 11,440 km) in the last year with a training volume of 106.3 km per week (SD 35.3 km/week, range 50 km to 200 km/week) . This reflects a much more extensive training and pre-race running experience compared to participants of normal (half-) marathon distances (for example, average weekly workload of 14 km of a cohort in a previous study on (half-)marathon runners ). This difference in training volume has also been reported by others  who stated that the emphasis during leisure training is usually more on speed, whereas ultramarathoners focus on duration and thus on endurance. Given their training workload, TEFR09 participants were extremely adapted to the demands of ultramarathon running. This is also supported by the observation that the participants' ultimate goal was completion of the whole multistage race rather than winning single stages. Accordingly, a rather low incidence of exercise-associated hyponatremia in ultramarathoners is reported [33, 57] and short term disturbances to the homeostasis of electrolytes or coagulation that may dominate during short race distances in less trained leisure athletes are rather unlikely to have contributed to the present results.
Given our above-mentioned criteria of substantiality and reversibility of present GM volume reductions only a subset of the discussed factors seem more likely than others to have contributed to present results. The loss of proteins as a likely relevant factor is further supported by the apparently strong common variation of body weight and GM volume which both returned to the baseline after 8 months. Furthermore, it is not unlikely that the return to baseline also aligned with hypercortisolism and possibly hyponatremia.
Therefore, further research is needed to find out each factor's contribution and their possible interaction leading to substantial and reversible GM volume loss during very long distance running.
Strengths and limits
The main strength of this study was its unique setting with a naturalistic and continuous observation of ultraendurance athletes reaching the limits of physical endurance. However, this setup entailed its own limitation since the number of participating athletes was small and attendance varied. Nevertheless, this is the first study to report brain volume changes or possible brain lesions of multistage ultramarathon runners observed during the race with a mobile MRI scanner. Since this truck-mounted scanner was available only for the run, scanning before the race and on follow-up had to be performed on different scanners although these MRI scanners were identical models and used identical sequence parameters. Due to the scanning on different scanners this report is limited to the exploration of global brain volumes, which are thought to be much less sensitive to the problem of using different MRI scanners than voxel based morphometric analyses of regional volume differences. Furthermore, a calculation of total intracranial volume has shown that measurements were nearly identical across the different scanners. Variations of intracranial volumes were statistically insignificant and with 0.3% of the total more than a magnitude less than observed gray matter volume changes of 6%. Also, the results are deemed plausible, since the measurements during the race were performed on the same mobile scanner and the volume decrease continues from timepoint 1 over 2 to 3. This is expected because of the protracted metabolic load during the run. Systematic errors due to the change of the scanner between timepoint 1 and 2 would affect only the first comparison.
Therefore, present results on GM volumes changes over time are thought to be robust, although they had to be acquired on different scanners.