Reply to: Hyperintense FLAIR signal in the anterior cranial fossa

replying to C. Graf et al. Nature Communications 10.1038/s41467-026-72196-z (2026)

We thank Graf et al. for their interest in our article, “Non-invasive MR imaging of human brain lymphatic networks with connections to cervical lymph nodes”¹, and for the opportunity to engage in this scientific dialogue. Their Matters Arising raises thoughtful technical points concerning our interpretation of the hyperintense signal observed in the anterior cranial fossa on 3D-T2 FLAIR MRI, which we proposed may correspond to ventral dural lymphatic structures.

We welcome the opportunity to clarify elements of our methodology and interpretation. While we appreciate the emphasis on careful imaging validation, we believe that some aspects of their critique may not fully reflect the methodological choices, physical considerations, and biological context that informed our original study. We address three central areas of clarification below.

Mischaracterization of our phantom study and imaging methodology

We recognize that the Graf et al. phantom provides a valuable conceptual demonstration of field inhomogeneity effects. While our interpretations differ, we appreciate the utility of such proof-of-principle models in encouraging further investigation into imaging artifacts in susceptibility-prone regions. Graf et al. base their critique on phantom and in vivo data comparing a standard adiabatic inversion pulse with a custom-engineered B₀- and B₁-robust inversion pulse1. In their setup, hyperintense signals in regions of known susceptibility variation, such as near a paramagnetic phantom cylinder or within the anterior cranial fossa, disappear when using their optimized pulse. Based on this, they infer that the FLAIR signal we reported is likely artifactual, stemming from incomplete inversion in B₀-inhomogeneous regions.

We respectfully offer a different interpretation. While we agree that MRI is inherently sensitive to field inhomogeneities and susceptible to artifacts under certain conditions, we believe the phantom model used by Graf et al. does not appropriately replicate the biological or imaging conditions of our study. Our protocol did not involve any exogenous contrast agents, neither gadolinium nor other susceptibility-inducing materials. Rather than testing field-related artifacts, our aim was to non-invasively detect a signal from endogenous interstitial fluid compartments enriched in macromolecules. To this end, we employed a 3D-T2 FLAIR sequence optimized to enhance slow-moving, protein-rich fluid while suppressing the free water signal.

To further clarify our perspective, we respectfully emphasize that our concerns are not aimed at the methodological validity of the Graf et al. phantom study itself, but rather at the limitations inherent in comparing such models directly with in vivo human imaging. In particular, the signal observed in our study was confined to the anterior cranial fossa, extending primarily along the anterior-posterior axis adjacent to the olfactory nerves and midline dura. We recognize, however, that anatomical confinement and reproducibility alone do not exclude susceptibility-related artifacts, especially in regions with complex geometry and air-tissue interfaces. In contrast, the artifacts reported in the Graf et al. phantom extended along multiple axes. We view this difference as highlighting the challenges of directly comparing simplified phantom models with in vivo human imaging, rather than as evidence distinguishing biological signal from artifact.

While their paramagnetic phantom is useful for evaluating susceptibility-driven artifact behavior, it does not replicate the dielectric, diffusional, or relaxation properties of CSF-adjacent dural spaces. In contrast, our phantom study was designed to simulate physiological conditions. We used non-paramagnetic, protein-enriched fluid (albumin), consistent with protein concentrations observed in interstitial and lymphatic compartments^2,3. We also adjusted our setup to approximate human body temperature, an approach explicitly described in our article, to further mimic in vivo conditions.

To rigorously evaluate whether inversion inefficiency alone could account for the observed signal, a phantom would need to be constructed with biologically relevant, non-susceptibility-based materials (e.g., albumin), and operated under physiologic pH, temperature, and relaxation parameters. We acknowledge that no phantom perfectly recapitulates in vivo microenvironments, but we believe our approach more closely mimics the biophysical context of the human brain’s interstitial and meningeal compartments than a purely susceptibility-driven model.

Vendor-specific pulse behavior and scientific scope of critique

We acknowledge the concern raised by Graf et al. that the hyperintense FLAIR signal we observed in the anterior cranial fossa could result from inversion inefficiency in B₀-inhomogeneous regions. While the effects of B₀ and B₁ field variation on inversion recovery are well known, their argument appears to assume uniform behavior across MRI systems and pulse designs, a notion that does not align with clinical or technical reality.

In practice, 3D FLAIR imaging performance, particularly regarding inversion robustness and fluid suppression, is strongly influenced by vendor-specific sequence implementations. Differences in RF pulse calibration, gradient strength, magnetization preparation modules, k-space trajectory, and interaction with the tissue dielectric environment all contribute to variable signal suppression, even when sequences are nominally equivalent.

Garf et al. used a Philips 3 T MRI system with a custom B₀-robust inversion pulse, while we employed a Siemens Prisma 3 T scanner using a clinically available SPACE-based 3D-T2 FLAIR sequence. These systems differ substantially in adiabatic pulse behavior, refocusing echo trains, and SAR constraints. Therefore, direct comparison between their phantom-based observations and our in vivo results is not methodologically appropriate.

FLAIR imaging is particularly sensitive for detecting proteinaceous fluids, such as in bacterial meningitis or other CSF-enriched conditions. However, sequence performance depends heavily on TR, TE, and inversion strategy. FLAIR implementations, whether conventional inversion recovery, T2-prep, or T2-selective pulses, differ in their sensitivity to macromolecular content. In our case, we used TR/TE/TI values of 5000/1800 ms on a Siemens Prisma scanner (Erlangen, Germany), while they reported TR/TI values of 8000/2400 ms on a Philips 3 T system (Best, The Netherlands), without disclosing TE. These parameter and vendor differences limit meaningful direct comparison.

In our phantom experiments, we found that inversion recovery and echo time are critical for differentiating protein concentrations. We also evaluated lymphatic CSF and interstitial fluid (ISF) in other protein-rich calvarial compartments. Notably, the protein concentration in the perilymphatic space (PLS) of guinea pigs is approximately 7.2 times higher than in the adjacent endolymphatic space (ELS)⁴. Fukutomi et al.⁵ demonstrated that conventional 3D-FLAIR inversion recovery could not distinguish between these compartments, but that T2-preparation pulses enabled this contrast enhancement even without gadolinium. These findings underscore how specific combinations of T2-prep timing and inversion delay can modulate FLAIR sensitivity to protein-rich compartments, an effect consistent with our observations. Viewed in this context, the signal attenuation reported by Graf et al. under B₀-robust pulse conditions may not necessarily indicate artifact, but could instead reflect a shift in contrast sensitivity resulting from pulse sequence design. This underscores the broader principle that signal acquisition strategies can influence visibility in 3D-FLAIR, and that selective suppression alone does not preclude potential biological relevance.

We similarly employed a T2-selective inversion recovery approach, as outlined in our protocol publication⁶, which was optimized for enhancing signal from macromolecule-enriched meningeal compartments. Our scan parameters, coil setup, and reconstruction pipeline were tuned to maximize contrast in these subtle fluid spaces and differ considerably from the methods used by Graf et al.

Given these differences in scanner platform, pulse design, and imaging strategy, signal discrepancies between studies are not unexpected. Rather, they reinforce the importance of interpreting results within the context of platform-specific sequence behavior. We appreciate the authors’ contribution to pulse optimization and welcome further discussion on refining lymphatic imaging techniques across vendors.

We do not intend to discount the observations of Graf et al. based solely on vendor differences. Rather, we emphasize that such technical disparities should inform, but not invalidate, comparative interpretation. As multi-vendor harmonization continues to improve, we agree that future collaborative efforts across imaging platforms will be essential to advance and validate lymphatic imaging strategies.

Biological relevance and reproducibility

We emphasize that our study was not intended as a pulse sequence development paper, but rather as a biologically motivated imaging investigation aimed at visualizing macromolecule-enriched interstitial compartments that may correspond to meningeal lymphatic structures. The signal patterns we observed were consistently reproducible across a large number of subjects and imaging planes, aligning with known anatomical trajectories from histological and tracer studies, particularly in the ventral cranial fossa, parasellar regions, and dural sinus–adjacent compartments. These signals were spatially discrete and anatomically patterned, and we interpret this distribution cautiously as compatible with plausible glymphatic and lymphatic-associated efflux zones, while emphasizing that susceptibility-related effects remain a credible alternative explanation that requires orthogonal validation.

We fully recognize the interpretive complexity of subtle 3D-FLAIR signals and the longstanding challenges associated with imaging meningeal lymphatic structures. The anterior cranial fossa, in particular, is a well-recognized artifact-prone region because of its proximity to air–tissue interfaces, complex bony anatomy, and local B₀ inhomogeneities. These factors can contribute to signal variability that may partially reflect artifacts. Nonetheless, the ventral signals we observed were anatomically coherent and reproducible across individuals, imaging planes, and sessions, suggesting a degree of structural consistency that merits further investigation. Accordingly, we support cautious interpretation of ventral FLAIR signal patterns and emphasize that MRI-based observations alone cannot establish biological identity. Clarifying the origins of these signals will require convergent approaches that combine advanced imaging with independent anatomical and histological validation.

Across both human and animal studies, imaging artifacts have frequently confounded the detection of lymphatic structures due to their small size, low flow rates, and limited contrast properties. The absence of definitive biological markers further complicates interpretation, and MRI, particularly in artifact-sensitive regions, remains susceptible to acquisition-dependent variability. In light of these challenges, we remain open to the possibility that a subset of the observed signals may reflect technical artifacts, and we cannot entirely rule this out. Moreover, brain lymphatic imaging has historically been difficult, largely due to the high artifact sensitivity of these delicate structures. Their small size, fluid content, and atypical architecture fall outside standard imaging norms, making non-invasive evaluation especially challenging. While MRI is a powerful tool, its sensitivity to subtle variations can complicate interpretation in both clinical and research contexts.

Given these limitations, we agree that it is not scientifically possible to assert that all observed signals are definitively artifact-free. However, the consistent spatial distribution and alignment with known lymphatic pathways offer meaningful support for their biological relevance. Future studies incorporating standardized imaging protocols and histopathological correlation will be essential to further clarify and validate these findings.

Conclusion

We appreciate the thoughtful and technically detailed feedback from Graf et al. and fully agree that sequence optimization and artifact recognition are critical for advancing neuroimaging. However, we believe their modeling approach and technical assumptions do not fully capture the biological, methodological, or vendor-specific context of our study. In our view, signal differences resulting from theoretical inversion enhancements do not negate the reproducibility, anatomical consistency, or physiological plausibility of the structures we identified.

We thank the editors for facilitating this scientific exchange and fostering constructive interdisciplinary dialogue. We strongly support continued collaboration among radiologists, physicists, and neuroscientists to refine imaging techniques and deepen our understanding of the meningeal and glymphatic systems.

Research involving human participants and/or animals

No new research involving human participants or animals was conducted as part of this reply.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.A. and O.A. jointly conceived and developed the response. Both authors contributed to the data interpretation and the writing and revision of the reply. Both authors reviewed and approved the final version of the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

This reply does not include new datasets. All data referenced or discussed are available in the original publication and associated supplementary materials. No additional datasets were generated or analyzed.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72195-0.