Voxel-based morphometry (VBM) is a widely used non-invasive imaging tool (1–3) to investigate patterns of structural brain change either between cohorts, e.g. various disease entities vs. healthy controls (4, 5) or longitudinal, e.g. before and after a specific learning task such as juggling or learning (6, 7). VBM essentially involves voxel-wise statistical analysis of pre-processed structural T₁-weighted MR images. Longitudinal VBM analyses provide the possibility to compare patterns of brain change over a distinct period.

VBM has been increasingly used to describe differences in brain structure between chronic pain patients and controls (8). A striking feature of all of these studies is the fact that gray matter changes (9) were not randomly distributed, but concerned defined and functionally highly specific brain areas – namely, areas involvement of supraspinal nociceptive processing (10). Most of these changes are discussed to represent damage, loss or even gain of brain gray matter, reinforcing the idea of measuring brain plasticity by T₁-weighted signal differences using VBM. An abundance of human studies, including several of our own, report morphologic alterations of the brain in areas responsible for the transmission of pain in patients suffering from chronic pain conditions such as phantom pain, chronic back pain, neuropathic pain, irritable bowel syndrome, fibromyalgia and two types of frequent headaches (6, 11–13). Several recent longitudinal studies involved learning paradigms and nearly all of them attribute their findings to be (a) specific for the respective learning paradigm and (b) to imply some kind of change on cell-level, mostly synaptogenesis (14, 15). However, at least three recent studies question the common belief that brain changes observed through VBM are always based on changes in cellular or neuronal structure of gray matter respectively. Tost et al. demonstrated GMV changes after single dose administration of a neuroleptic drug, suggesting interpretation of these morphological short-term changes as synaptogenic alterations (16). More recently, Franklin et al. identified short-term effects of a single dose of medication (baclofen) and demonstrated a partial overlap of these changes with concurrent changes in perfusion f-MRI (3). The authors concluded that T₁-relaxation time for arterial blood and gray matter might not be clearly distinguishable and therefore structural findings could be at least partially attributed to changes in blood flow suggesting that the observed changes have to be interpreted as T₁-weighted signal differences and not as gray matter alterations (3). Tardif et at. demonstrated that increased blood volume from vasodilation during hypercapnia, induced by hyperventilation, is associated with an overestimation of cortical thickness (1.85%) and gray matter volume (3.32%), and that both changes in O2 concentration and blood volume [i.e. cerebral blood flow (CBF)] lead to changes in the T₁ value of tissue (17).

As the central mechanisms of pain processing are well understood in experimental as well as chronic pain, we chose pain stimulation in healthy participants as a model to study in vivo non-pharmacological short-term morphometric changes in humans. In this study we primarily hypothesized that electrical high intensity pain stimulation could lead to ultra-short term pain dependent changes in brain structure using high-field magnetic resonance imaging (MRI). Only upon a closer look at the results obtained from a control group without any electrical painful stimulation we changed the study design and further developed our hypothesis toward scanner environment or loco-regional blood flow fluctuations being responsible for the detected GMV changes.

A detailed overview of the scanning sessions including the 4 projects (A-D) can be found in Figure 1. Projects A, B, and C were scanned in Hamburg, Germany on a Siemens® 3T TRIO Scanner. Data of project D was obtained at Innsbruck (Siemens® 3T VERIO). Participants of project A & B are identical. Participants of project C & D represent new and independent groups.

All subjects were right-handed and male. They were recruited locally and all systematically screened by an experienced full-time neurologist and pain specialist to ensure they were free of any lifetime history of neurological or psychiatric illness. Subjects with acute or chronic pain related disorders were excluded. Consumption of any medication (including pain medication) seven days prior to scanning led to exclusion. All participants had normal or corrected-to-normal vision and gave written informed consent following a detailed explanation of study procedures. The study was approved by the local ethics committee (Ärztekammer Hamburg PV3562) and all participants gave written informed consent.

In total 72 healthy volunteers (in 96 scanning sessions) were consecutively enrolled in the study that comprised 4 scanning sessions (Projects A-D, Figure 1). Because we observed more statistical power and larger effects in the sample of project A (n = 31), we always used randomly selected 20 volunteers in each sample for the structural data to obtain a balanced design in terms of sample size and statistical power. We therefore only report the results for this balanced design that allows a fair and reliable comparison between all experiments without bias due to different sample sizes.

In none of these projects we found a significant difference in the values of total intracranial volume or total gray matter volume between sessions.

Our main finding is an exceptionally robust, yet locally distinctive decrease in local gray matter T1-signal after short-term repetitive MPRAGE imaging in healthy volunteers. These changes were mostly located in the anterior and rostral cingulate cortex and both insula cortices. This finding could be reproduced in 3 different cohorts, including 96 scanning sessions and 72 healthy volunteers on two different 3 Tesla scanners. Additionally, we also identified increases in local gray matter volume between subsequent scans. Again, these effects were stable over the different cohorts, but included areas were more threshold dependent, compared to the found decreases in GMV.

Opposed to the schedule of longitudinal studies, which usually comprise several weeks or even months our study focuses on short-term repetitive MPRAGE acquisition. Our finding resulted from a repetitive MPRAGE image acquisition time frame of just approximately 30 min separated by either a simple 15 min pain task (project A) or “no specific task” (projects B-D) in healthy volunteers. We originally interpreted these changes to be caused by, or secondary to, the pain stimulation in the first cohort (project A). The reason for this was that these changes matched perfectly with BOLD fMRI data under pain stimulation (20, 21) and the anatomical location of these changes was within distinct central pain processing areas (22) (Figure 3). Much to our surprise we observed the same effect in the post-hoc scanned control cohorts (projects C and D) with identical scanning protocol but without any pain stimulation. This finding argues strongly against pain being causative for the observed changes. The observed results could only be detected within a specific scanning session if an EPI or REST (i.e. independent of a stimulus presentation) sequence was run before the second T₁-weighted image. The effect was however more pronounced in EPI sequenced sessions than in the rest sessions, which suggests that future morphometric studies that are designed as T₁-weighted images in combination with a functional MRI scanning session should focus on a consistent timing at which time-point during the experiment the T₁-weighted scan (before or after the EPI sequences session) is conducted.

While temporal instabilities of the MR systems used as the source of the observations cannot be definitely excluded, this seems rather unlikely for several reasons: Firstly, two different MR systems with different gradient hardware and at different sites were involved. Second, most system instabilities, e.g. due to gradient and shim heating, are related to, or at least more pronounced for, severe hardware demands like EPI measurements, but in the present study differences were also observed without any measurements in-between. Third, the regional effects reported are quite localized while the scanner hardware (RF chain, gradient and shim hardware) presumably would cause more global changes.

Due to the short time frame a biological interpretation of our results is challenging. Possible explanations could be hemodynamic changes in blood flow (23) and increased blood pressure (17, 24) caused by the subjects' perception in the scanner. Data from neuropsychological studies investigating the effect of anxiety disorders on gray matter changes further support this assumption (25–27). Other reasonable explanations comprise possible fluid shift between extra- and intracellular spaces. When taking the relatively low specific absorption rate (SAR) into account, warming induced by radio-frequency electromagnetic fields utilized to acquire images (28) can be discarded as a possible explanation. We also conclude that the rating sequences applied could not result in the GMV changes we found as they are distributed specifically in brain networks associated with pain. Biological effects other than thermal caused by radiofrequency electromagnetic fields are complex and still highly speculative (29). Whether the respective regions such as anterior cingulum and insula are especially susceptible for magnetic or radiofrequency influence therefore explaining our results should be addressed in future studies. We found no literature on the interaction of specific absorption rate (SAR) on findings using voxel based morphometry or statistical parametric mapping (SPM).

However, the potential influence of the magnetic fields (static and time-varying) on brain function has been discussed since the late nineties (30). One might speculate that the rather local signal effects in our study may be a result of the geometrical distribution of the neuronal tracts in the various cortical brain regions being differently susceptible to the rapidly changing magnetic fields. There is some evidence that in another technique using rapidly changing magnetic fields, i.e. transcranial magnetic stimulation (TMS), the effects on neuronal activity are orientation dependent (31).

A decrease in gray matter estimations between the first and the subsequent conditions can be interpreted as a decline from baseline or alternatively as a return to baseline after prior signal enhancement. One could speculate that the changes that we see between scan S_HH1 vs. scan S_HH2 after noxious input are triggered by expectation, i.e. we actually describe an apparent “activation” (in the broader sense of a biological signal) or “increase” in brain areas before the experiment as volunteers anticipated the pain paradigm. However, when comparing the scan (project A, S_HH1) before the experiment with the scan of the post-hoc experiment (project B, S_HH3) (involving no expectation as the volunteers were told that no noxious input would follow), no significant change in the GMV at the defined thresholds could be observed. Our interpretation is that the respective changes of the T₁-signal are reversible and widely rules out a putative bias introduced by expectation or conditioning that could have influenced our results.

We are fully aware that the results of our study have to be interpreted with great caution. What gives strength to our data is the fact that the results were very robust and could be reproduced in 3 independent projects (A/C/D) including almost 100 scanning sessions and survived very conservative statistical correction (FWE, p < 0.05) in all analyses. As the protocols were done on different scanner models using different head coils we can for the most part rule out a probable scanner-specific bias. Moreover, the preprocessing is using an inverse-consistent registration approach that guarantees independence from the order of selecting the time points. This was an issue of older approaches where the baseline image was used as reference for the registration and therefore no interpolation was necessary for this baseline image, which led to unbalanced interpolation effects between the time points. In order to address and test that our longitudinal preprocessing is not dependent on the order of time-points we changed the selection order for data from Experiment D and found no dependency from this order. This further demonstrates the inverse-consistency of our preprocessing. Additionally, we tested the global conjunction across all three experiments using a minimum statistic in SPM12, which resulted in similar regions that were found for the three experiments separately. Although the extent and the location of the findings differ between the experiments, we could find a pattern that is somewhat consistent across all experiments. Moreover, a similar pattern in the global conjunction analysis could also be observed when we additionally used two different VBM pipelines of SPM12 and FSL6. This emphasizes that the observed effects are not a simple artifact due to the processing pipeline used.

Whatever the biological/methodological interpretation underlying the GMV change, we note that these reductions were definitely not randomly distributed but only observed in specific areas and are exceptionally robust even over different cohorts and different scanners.

We note that this may have serious implications for published and future studies especially in (but not limited to) those that are designed as T₁-weighted images in combination with a functional MRI scanning session. We propose that all future studies investigating T₁-signal changes between cohorts but especially in longitudinal experimental designs (i.e. “brain plasticity studies”) should focus on and specifically report a consistent timing at which time-point during the experiment the T₁-weighted scan is conducted. Additionally, this study also addresses the imperative necessity of a control group (with identical timing parameters) for longitudinal imaging studies to monitor for a possible systematic scanner – time effect influencing T₁-signal.

The current study suggests that repetitive short-term MPRAGE imaging can show significant T₁-signal changes resulting in altered GMV estimations in highly specific and circumscribed brain areas. Although we can only speculate about the underlying mechanisms, the robustness of the results is striking. We would like to draw attention to the necessity of a careful order and timing of MR protocols and paradigms in longitudinal VBM studies.

Study data supporting the findings of this study are available from the corresponding author upon reasonable request after ethics approval and receipt of a signed material transfer agreement.

The studies involving human participants were reviewed and approved by Ärztekammer Hamburg PV3562. The patients/participants provided their written informed consent to participate in this study.

GB: data collection, data analysis, data interpretation, and writing the manuscript. IE: data collection, data interpretation, and revising the manuscript. MM and CG: data analysis, data interpretation, and revising the manuscript. FF and MV: data collection, data analysis, and revising the manuscript. AM: conception and design of the study, data interpretation, and revising the manuscript. All authors contributed to the article and approved the submitted version.

This work was supported by the German Research Foundation, SFB936/A5 (AM).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.