A molecular array for 10-second diagnosis of common spinal tumor types with picosecond infrared laser mass spectrometry

Abstract

Background: Improving the surgical outcomes for commonly occurring spinal neoplasms of extradural and intradural extramedullary origins requires precise intraoperative diagnosis provided by highly trained neuropathologists.

Methods: Through a retrospective study of n = 319 patient specimens, verified where appropriate by learning curve analysis to be sufficient for statistically significant observations, we aimed to assess the utility of 10-second picosecond infrared laser mass spectrometry (MS; PIRL-MS) for non-subjective diagnosis of major spinal tumor types of metastatic carcinoma, schwannoma, and meningiomas.

Results: The sensitivity and specificity values of spinal tumor-type diagnosis (based on n = 182 independent specimens) were (93% ± 1)% and (97% ± 2)%, respectively. This classification utilizes n = 41 cellular lipids including phosphatidylcholines, sphingomyelins, phosphatidylethanolamines, and ceramides whose identities were established using high-resolution tandem MS. Furthermore, the accuracy of diagnosis of a model that contained n = 97 meningioma and n = 106 schwannoma was not drastically influenced by the presence of n = 54 additional intradural extramedullary spinal neoplasms of myxopapillary ependymoma, neurofibroma, paraganglioma, and solitary fibrous tumor types in the differential diagnosis, confirming the generalizability and robustness of the identified molecular array in rendering correct classification even in the presence of data not seen previously by the model.

Conclusions: The identified lipids form a "molecular array" for robust diagnosis of meningioma and schwannoma tumors by non-pathologists in a manner similar to genomic, transcriptomic, or methylomic arrays used to diagnose brain cancer types, albeit on a much faster timescale of seconds as opposed to hours.

Keywords: diagnostic biomarkers; lipidomics; mass spectrometry; non-subjective diagnosis; picosecond infrared laser mass spectrometry; rapid diagnosis; spinal tumors.

Figure 1.

Representative 10-second PIRL-MS spectra of meningioma, schwannoma, and metastatic cancers. In this figure, we are providing the representative PIRL-MS spectra of the major types of spinal tumors alongside the total ion count (TIC) values for these spectra. Here, we have highlighted the abundant m/z values in black font. As can be seen here, PIRL-MS profiling across the mass range of 100–1000 Da (Daltons) reveals distinct molecular profiles for these 3 tumor types. The n = 41 ions important for discrimination (Table 2) were visible in these representative spectra (highlighted in red font / underlined). A qualitative comparison of Figure 1 and Table 2 results with desorption electrospray ionization mass spectrometry (DESI-MS) spectra of meningiomassuggests complementary molecular information between PIRL-MS and DESI-MS as previously reported using murine models of human medulloblastoma brain cancers.

Figure 2.

Multivariate PCA-LDA modeling of major spinal tumor types. Here we show 2 molecular models built from n = 122 independent patient specimens (n = 39 meningiomas, n = 41 schwannoma, and n = 42 metastatic tumors) generating the n = 959 sampling events illustrated in the plot. (A) PCA-LDA model for true annotations wherein we have grouped each tumor type into their correct pathology class as verified by a neuropathologist. (B) A permutated PCA-LDA model bearing mixed class annotations wherein each group (class) is comprised of ~ equal numbers of sampling events from all other participating classes. As can be seen, the true annotation model (panel A) exhibits a clustering of each tumor type data, supported by cross-validation results (Table S2) resulting in accuracies of 98.84% (for 20% leave-out) and 96.27% (for full group leave-out). The permutated model of Panel B on the other hand does not show any group separations, consistent with its poor cross-validation statistics (Table S3) of 41.32% (for 20% leave-out) and 35.54% (for full group leave-out).

Figure 3.

PCA-LDA models for additional spinal tumor types in the differential diagnosis. To generate this figure, we subjected n = 2148, PIRL-MS spectra from n = 257 specimens to PCA-LDA modeling: meningioma (n = 97), schwannoma (n = 106), myxopapillary ependymoma (n = 18), neurofibroma (n = 18), paraganglioma (n = 9), solitary fibrous tumors also known as hemangiopericytoma (n = 9). The cross-validation (20% and full-group leave-out) statistics for these 2 PCA-LDA models are provided in Table S6 (98.75% and 96.38%, respectively), suggesting separation between classes. The confusion matrix in Table S6 suggests minimal misclassification between meningioma and schwannoma is taking place with the additional intradural extramedullary tumors included. For clarity, 2 views of the same model are shown. It must be noted that the 3-dimensional projections shown here do not fully represent the class separations seen in other dimensions. Therefore, the cross-validation results of Table S6 must be consulted.