Over 20% of cancer patients develop secondary brain tumours, or metastases. Current gold-standard MRI diagnosis relies upon gadolinium enhancement on a T1-weighted scan, indicative of a permeable blood-brain barrier. This is a late-stage event and thus, new methods enabling earlier diagnosis are urgently needed. We have previously shown that it is possible to discriminate between different inflammatory lesions in the CNS in rats¹, as well as between different stages of multiple sclerosis in patients², through NMR biofluid (blood/urine) metabolomics. We believe that this ability is due, at least in part, to alterations in CNS metabolism, which provoke a specific metabolic signature in the biofluids studied. It is known that tumour metabolism differs markedly from normal brain and that brain metabolism itself will be altered by tumour presence. On the basis of the above, therefore, we hypothesised that the presence of brain metastases could be detected, in vivo, through NMR analysis of biofluids.Metastatic mammary carcinoma cells (murine 4T1-GFP) were injected into BALB/c mice, via intracerebral, intracardial or intravenous routes to induce differing cerebral and systemic tumour burdens. Urine was collected at days 0 and 10 from all animals, and at days 5, 21 and 35 in animals injected intracerebrally. Samples from naïve, day 0 and vehicle-injected mice served as combined control cohorts. ¹H-NMR NOESY pre-saturation spectra were acquired for each sample using a Bruker Avance III spectrometer equipped with a ¹H TCI cryoprobe. Two-dimensional COSY ¹H-NMR spectra were acquired from a single sample within each group to assist with metabolite identification. Statistical pattern recognition and modelling was applied to identify spectral differences and to identify commonalities indicative of brain metastasis burden. A robust method using unknown samples was used to validate model predictions. Additional models using alternate cell lines were used for further validation (B16F10 murine melanoma in BL6 mice and MDA231BR-GFP human breast carcinoma in SCID mice).

Significant differences in ¹H NMR spectra were found between control cohorts and animals with tumour burdens at all timepoints for the intracerebral 4T1-GFP metastasis model (q² = 0.54, 0.58, 0.81 and 0.80 for days 5, 10, 21 and 35, shown in Fig. 1A, B, C and D respectively; q² > 0.4 considered significant).

Models became stronger, with higher sensitivity and specificity, as the timecourse progressed indicating a more severe tumour burden. Sensitivity and specificity for predicting a blinded testing set were 0.89 and 0.82, respectively, at day 5, but both rose to 1.00 at day 35. Key metabolites driving the separations were identified and quantified relative to control (Fig. 2).

Significant separations were also found between control and day 10 animals for all 4T1-GFP injected mice irrespective of route (q² = 0.70, 0.63 and 0.78 for intracerebral, intracardiac and intravenous routes, respectively). The metabolites underpinning separations in each case differed indicating differentiation between systemic and CNS metastatic burden, but with common patterns that suggest a “fingerprint” for brain metastasis.Our data indicate that animals with brain metastases can be identified using NMR spectra of urinary metabolic profiles, and differentiated from animals with a predominantly systemic metastasis burden. These models were highly sensitive and specific at predicting blinded testing sets. This approach may identify a set of biomarkers for the diagnosis of brain metastasis earlier than is currently possible, and may provide insight into metabolic disruption in brain metastasis.Urine metabolomics based upon ¹H-NMR spectroscopy is sensitive and specific for both following tumour progression and distinguishing animals with heavy brain burdens of tumours relative to heavy systemic burdens.