Precise co-registration of mass spectrometry imaging, histology, and laser microdissection-based omics

Mass spectrometry imaging (MSI) is an analytical technique for the unlabeled and multiplex imaging of molecules in biological tissue sections. It therefore enables the spatial and molecular annotations of tissues complementary to histology. It has already been shown that MSI can guide subsequent material isolation technologies such as laser microdissection (LMD) to enable a more in-depth molecular characterization of MSI-highlighted tissue regions. However, with MSI now reaching spatial resolutions at the single-cell scale, there is a need for a precise co-registration between MSI and the LMD. As proof-of-principle, MSI of lipids was performed on a breast cancer tissue followed by a segmentation of the data to detect molecularly distinct segments within its tumor areas. After image processing of the segmentation results, the coordinates of the MSI-detected segments were passed to the LMD system by three co-registration steps. The errors of each co-registration step were quantified and the total error was found to be less than 13 μm. With this link established, MSI data can now accurately guide LMD to excise MSI-defined regions of interest for subsequent extract-based analyses. In our example, the excised tissue material was then subjected to ultrasensitive microproteomics in order to determine predominant molecular mechanisms in each of the MSI-highlighted intratumor segments. This work shows how the strengths of MSI, histology, and extract-based omics can be combined to enable a more comprehensive molecular characterization of in situ biological processes. Electronic supplementary material The online version of this article (10.1007/s00216-019-01983-z) contains supplementary material, which is available to authorized users.

Step-by-step description of the co-registration between mass spectrometry imaging, histology, and the laser microdissection system.  identified and quantified using the label-free quantification (LFQ). These following settings were applied: Uniprot reviewed human database, trypsin digestion with two maximum missed cleavage sites, methionine oxidation as variable modification and carbamidomethyl cysteine as fixed modification, a minimal peptide length of seven amino acids, at least two peptides per protein (of which at least one is unique), and a maximum false discovery rate of 1%. The label-free intensities were normalized using the MaxLFQ algorithm.

Fig. S1
Co-registration of MSI to the high-resolution optical image. (a) The co-registration between the MSI data and the optical image was done via an affine geometric transformation between three landmarks of the laser shots in the matrix, that are visible in the high-resolution optical image (a, left column), and the corresponding MSI pixels (a, right column). (b) The estimation of the co-registration error was assessed by counting the number of pixels (pixel size ≈ 2 µm) for x and y in the optical image separating the center of the laser shot landmark from the corresponding MSI pixel plotted on top of the optical image

Fig. S2
Estimation of the LMD co-registration error at 5x magnification. The co-registration error between optical image and the LMD was determined by creating virtual shapes in the optical image with a known distance to nearby Tipp-Ex spots. These shapes were then cut by the laser of the LMD at 5x magnification. The co-registration error was then evaluated at 10x magnification by measuring the distance between the Tipp-Ex spot and the laser landmark in the tissue of the cut shapes

Fig. S3
Smoothing of the NNMF segmentation image. Image processing was performed on the raw segmentation results (left image) to increase the viability of microdissection. This included at first a smoothing of the segmentation image using the imopen Matlab function with a 2x2 square as structuring element. The result is shown in the right image

Fig. S4
Image processing after NNMF segmentation. Several steps of image processing on the smoothed segmentation image were further needed in order to detect the boundaries of each clusters (left to right columns). First, the smoothed segmentation image was divided into three binary images, each depicting the pixels belonging to one of the clusters (leftmost column). Each of these was further processed individually. Impurities were removed by deleting small areas (≤30 pixels in the 4-connected neighborhood) using bwareaopen (second column) and by filling holes in the 8-connected neighborhood using imfill (third column). The individual binary images were then warped to the dimensions of the histological image using imwarp with nearest-pixel interpolation. Once up-scaled, the last step of the image processing was to detect the external boundaries of all segments belonging to each cluster using bwboundaries (rightmost column)

Fig. S5
Microproteomics characterization of MSI-defined intra-tumoral clusters. LMD was performed for each of the MSI-defined intra-tumoral clusters by cutting out 0.3mm 2equivalent material for each MSI-cluster for microproteomics analysis. This resulted in the identification and label-free quantification of over 1400 common proteins. (a) Hierarchical clustering was performed on the log2-tranformed and z-scored data (1040 proteins remained) to group MSI-clusters and proteins by expressing similarity. (b) Cluster exclusive over-and under-expressed proteins (z-scores ≥+1 and ≤-1, respectively) were submitted to gene ontology analysis to determine differences in predominant molecular function between the three MSI clusters, and therefore between the different breast cancer subpopulations