I. INTRODUCTION

Section:

ChooseTop of pageABSTRACTI. INTRODUCTION <<II. EXPERIMENTIII. RESULTSIV. DISCUSSIONV. SUMMARY AND CONCLUSION...AUTHORS’ CONTRIBUTIONSREFERENCESA biomolecule’s distribution within a cell yields insight into the cellular processes that it influences. For this reason, many techniques have been developed to image the distributions of molecules of interest in cells. Though numerous approaches are currently available for visualizing specific proteins in cells, fewer techniques enable imaging nonprotein metabolites, such as lipids and sterols. One such technique is secondary ion mass spectrometry (SIMS).1–91. H. A. Klitzing, P. K. Weber, and M. L. Kraft, Methods in Molecular Biology: Nanoimaging Methods and Protocols, edited by A. A. Sousa and M. J. Kruhlak (Humana, Totowa, NJ, 2013), Vol. 950, pp. 483–501.2. B. L. Gorman and M. L. Kraft, Anal. Chem. 92, 1645 (2020). https://doi.org/10.1021/acs.analchem.9b044923. B. L. Gorman, A. N. Yeager, C. E. Chini, and M. L. Kraft, Characterization of Biological Membranes Structure and Dynamics, edited by M.-P. Nieh, F. A. Heberle, and J. Katsaras (De Gruyter, Berlin, 2019), pp. 287–322.4. M. L. Kraft and H. A. Klitzing, Biochim. Biophys. Acta 1841, 1108 (2014). https://doi.org/10.1016/j.bbalip.2014.03.0035. B. Johansson, Surf. Interface Anal. 38, 1401 (2006). https://doi.org/10.1002/sia.23616. D. R. N. Vos, S. R. Ellis, B. Balluff, and R. M. A. Heeren, Mol. Imaging Biol. 23, 149 (2021). https://doi.org/10.1007/s11307-020-01541-57. M. K. Passarelli and N. Winograd, Biochim. Biophys. Acta 1811, 976 (2011). https://doi.org/10.1016/j.bbalip.2011.05.0078. A. S. Mohammadi, X. Li, and A. G. Ewing, Anal. Chem. 90, 8509 (2018). https://doi.org/10.1021/acs.analchem.8b013959. J. S. Fletcher and J. C. Vickerman, Anal. Chem. 85, 610 (2013). https://doi.org/10.1021/ac303088m SIMS images of the surfaces of cells have revealed that some lipid species are heterogeneously distributed within the cell membrane.10–1610. S. G. Ostrowski, C. T. Van Bell, N. Winograd, and A. G. Ewing, Science 305, 71 (2004). https://doi.org/10.1126/science.109979111. M. E. Kurczy, P. D. Piehowski, C. T. Van Bell, M. L. Heien, N. Winograd, and A. G. Ewing, Proc. Natl. Acad. Sci. U.S.A. 107, 2751 (2010). https://doi.org/10.1073/pnas.090810110712. R. L. Wilson, J. F. Frisz, H. A. Klitzing, J. Zimmerberg, P. K. Weber, and M. L. Kraft, Biophys. J. 108, 1652 (2015). https://doi.org/10.1016/j.bpj.2015.02.02613. J. F. Frisz, H. A. Klitzing, K. Lou, I. D. Hutcheon, P. K. Weber, J. Zimmerberg, and M. L. Kraft, J. Biol. Chem. 288, 16855 (2013). https://doi.org/10.1074/jbc.M113.47320714. J. F. Frisz et al., Proc. Natl. Acad. Sci. U.S.A. 110, E613 (2013). https://doi.org/10.1073/pnas.121658511015. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biochim. Biophys. Acta 1860, 2004 (2018). https://doi.org/10.1016/j.bbamem.2018.04.00816. C. He et al., Proc. Natl. Acad. Sci. U.SA. 114, 2000 (2017). https://doi.org/10.1073/pnas.1621432114 For example, decreases in phosphatidylcholine and sphingomyelin have been detected in Tetrahymena thermophila following structural changes in the membrane.10,1110. S. G. Ostrowski, C. T. Van Bell, N. Winograd, and A. G. Ewing, Science 305, 71 (2004). https://doi.org/10.1126/science.109979111. M. E. Kurczy, P. D. Piehowski, C. T. Van Bell, M. L. Heien, N. Winograd, and A. G. Ewing, Proc. Natl. Acad. Sci. U.S.A. 107, 2751 (2010). https://doi.org/10.1073/pnas.0908101107 Additionally, our use of a NanoSIMS for imaging metabolically incorporated isotope-labeled lipids on the surfaces of mammalian cells has revealed localized regions of sphingolipid enrichment and a fairly even distribution of cholesterol within the plasma membrane.12–1512. R. L. Wilson, J. F. Frisz, H. A. Klitzing, J. Zimmerberg, P. K. Weber, and M. L. Kraft, Biophys. J. 108, 1652 (2015). https://doi.org/10.1016/j.bpj.2015.02.02613. J. F. Frisz, H. A. Klitzing, K. Lou, I. D. Hutcheon, P. K. Weber, J. Zimmerberg, and M. L. Kraft, J. Biol. Chem. 288, 16855 (2013). https://doi.org/10.1074/jbc.M113.47320714. J. F. Frisz et al., Proc. Natl. Acad. Sci. U.S.A. 110, E613 (2013). https://doi.org/10.1073/pnas.121658511015. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biochim. Biophys. Acta 1860, 2004 (2018). https://doi.org/10.1016/j.bbamem.2018.04.008Intracellular lipid distribution and the subcellular localization of other components have also been probed by performing SIMS in a depth profiling mode.17–2717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.493968118. D. Breitenstein, C. E. Rommel, R. Möllers, J. Wegener, and B. Hagenhoff, Angew. Chem. Int. Ed. 46, 5332 (2007). https://doi.org/10.1002/anie.20060446819. M. A. Robinson, D. J. Graham, and D. G. Castner, Anal. Chem. 84, 4880 (2012). https://doi.org/10.1021/ac300480g20. J. S. Fletcher, N. P. Lockyer, S. Vaidyanathan, and J. C. Vickerman, Anal. Chem. 79, 2199 (2007). https://doi.org/10.1021/ac061370u21. J. S. Fletcher, Biointerphases 10, 018902 (2015). https://doi.org/10.1116/1.490772722. A. Thomen et al., ACS Nano 14, 4316 (2020). https://doi.org/10.1021/acsnano.9b0980423. C. E. Chini, G. L. Fisher, B. Johnson, M. M. Tamkun, and M. L. Kraft, Biointerphases 13, 03B409 (2018). https://doi.org/10.1116/1.501973624. J. Brison, D. S. W. Benoit, S. Muramoto, M. Robinson, P. S. Stayton, and D. G. Castner, Surf. Interface Anal. 43, 354 (2011). https://doi.org/10.1002/sia.341525. D. J. Graham, J. T. Wilson, J. J. Lai, P. S. Stayton, and D. G. Castner, Biointerphases 11, 02A304 (2016). https://doi.org/10.1116/1.493479526. C. Szakal, K. Narayan, J. Fu, J. Lefman, and S. Subramaniam, Anal. Chem. 83, 1207 (2011). https://doi.org/10.1021/ac103060727. H. Tian, L. J. Sparvero, P. Blenkinsopp, A. A. Amoscato, S. C. Watkins, H. Bayır, V. E. Kagan, and N. Winograd, Angew. Chem. Int. Ed. 58, 3156 (2019). https://doi.org/10.1002/anie.201814256 When using a NanoSIMS instrument, depth profiling involves analyzing the same sample location multiple times.2,3,28,292. B. L. Gorman and M. L. Kraft, Anal. Chem. 92, 1645 (2020). https://doi.org/10.1021/acs.analchem.9b044923. B. L. Gorman, A. N. Yeager, C. E. Chini, and M. L. Kraft, Characterization of Biological Membranes Structure and Dynamics, edited by M.-P. Nieh, F. A. Heberle, and J. Katsaras (De Gruyter, Berlin, 2019), pp. 287–322.28. G. McMahon, B. J. Glassner, and C. P. Lechene, Appl. Surf. Sci. 252, 6895 (2006). https://doi.org/10.1016/j.apsusc.2006.02.17029. J. Nunez, R. Renslow, J. B. Cliff, and C. R. Anderton, Biointerphases 13, 03B301 (2018). https://doi.org/10.1116/1.4993628 Because the primary ion beam removes the top layer of material from the sample each time it is scanned across its surface, each image shows the composition slightly deeper in the cell than the one before it. The resulting series of NanoSIMS images are stacked and aligned to form a three-dimensional (3D) rendering that reveals the intracellular distribution of components of interest, such as lipids.1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681Although the individual 2D SIMS images acquired during depth profiling are easily converted into a 3D rendering, SIMS images lack topographic information, so the resulting 3D rendering shows a cell with a distorted morphology. Specifically, the curved top surface of the cell becomes flat, and the planar bottom surface of the cell becomes curved in the 3D rendering. Additionally, the secondary ion signals detected within the cell are positioned relative to the top surface of the cell in the rendering, whereas the true positions of the molecules that produced these signals should be relative to the underlying substrate. Thus, the 3D image stack must be reshaped so that it has the same morphology as the original sample. One strategy relies on using the height data encoded in correlated atomic force microscopy (AFM) or scanning probe microscopy (SPM) images to reshape the depth profiling data.30–3530. A. Wucher, J. Cheng, and N. Winograd, Anal. Chem. 79, 5529 (2007). https://doi.org/10.1021/ac070692a31. A. Wucher, J. Cheng, L. Zheng, D. Willingham, and N. Winograd, Appl. Surf. Sci. 255, 984 (2008). https://doi.org/10.1016/j.apsusc.2008.05.24632. T. Terlier, J. Lee, K. Lee, and Y. Lee, Anal. Chem. 90, 1701 (2018). https://doi.org/10.1021/acs.analchem.7b0343133. S. Jung, N. Lee, M. Choi, J. Lee, E. Cho, and M. Joo, Appl. Surf. Sci. 432, 90 (2018). https://doi.org/10.1016/j.apsusc.2017.05.21734. M. A. Moreno, I. Mouton, N. Chevalier, J.-P. Barnes, F. Bassani, and B. Gautier, J. Vac. Sci. Technol. B 36, 03F122 (2018). https://doi.org/10.1116/1.501946435. V. Spampinato, M. Dialameh, A. Franquet, C. Fleischmann, T. Conard, P. van der Heide, and W. Vandervorst, Anal. Chem. 92, 11413 (2020). https://doi.org/10.1021/acs.analchem.0c02406 Though this produces an accurate topography, correlated AFM or SPM imaging and NanoSIMS depth profiling analysis of cells is challenging to acquire due to the difficulty of predicting which cells will be conducive to SIMS depth profiling. An alternative strategy is to use the secondary ions produced by the substrate to define the bottom of the cell, which is then used to reshape the 3D image stack.18,19,25,3218. D. Breitenstein, C. E. Rommel, R. Möllers, J. Wegener, and B. Hagenhoff, Angew. Chem. Int. Ed. 46, 5332 (2007). https://doi.org/10.1002/anie.20060446819. M. A. Robinson, D. J. Graham, and D. G. Castner, Anal. Chem. 84, 4880 (2012). https://doi.org/10.1021/ac300480g25. D. J. Graham, J. T. Wilson, J. J. Lai, P. S. Stayton, and D. G. Castner, Biointerphases 11, 02A304 (2016). https://doi.org/10.1116/1.493479532. T. Terlier, J. Lee, K. Lee, and Y. Lee, Anal. Chem. 90, 1701 (2018). https://doi.org/10.1021/acs.analchem.7b03431 However, sputtering through an entire mammalian cell in order to detect the substrate ions beneath it is extremely time consuming when depth profiling is performed on a NanoSIMS with a small-diameter primary ion beam (>55 h would be required for a sputter rate of 90 nm/h and cell height of 5 μm). In addition, when depth profiling is performed on a NanoSIMS 50 (or 50 L) instrument, a detector may not be available for the collection of substrate-specific ions because all five (or seven) of detectors may already be collecting component-specific secondary ions.We sought to develop an approach for depth correcting 3D SIMS images that does not require correlated topography measurements or the detection of the substrate. We hypothesized that we could use the secondary electron images that are collected in parallel with the negatively charged secondary ions for depth correction because the contrast present in these secondary electron images is a visual cue for depth perception.36,3736. S. Ichihara, N. Kitagawa, and H. Akutsu, Perception 36, 686 (2007). https://doi.org/10.1068/p569637. E. Kaplan, The Senses A Comprehensive Reference, edited by R. H. Masland, T. D. Albright, T. D. Albright, R. H. Masland, P. Dallos, D. Oertel, S. Firestein, G. K. Beauchamp, M. Catherine Bushnell, A. I. Basbaum, J. H. Kaas, and E. P. Gardner (Academic, New York, 2008), pp. 29–43. In this study, we used our previously published NanoSIMS depth profiling data of an Madin–Darby canine kidney (MDCK) cell1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681 to explore whether the pixel intensities in the secondary electron images that were acquired during NanoSIMS analysis could be used to reconstruct the morphology of the cell at the analysis location. We also assessed whether the evolution of the cell’s morphology during depth profiling suggested variations in the sputter rate. Our results indicate the depth corrected 3D SIMS images produced with our strategy better match the actual cell morphology than the uncorrected 3D images, and these more accurate 3D representations may facilitate the identification of subcellular features.

II. EXPERIMENT

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. EXPERIMENT <<III. RESULTSIV. DISCUSSIONV. SUMMARY AND CONCLUSION...AUTHORS’ CONTRIBUTIONSREFERENCES

A. Sample preparation and SIMS analysis

The SIMS data used in this report had been previously reported.1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681 Briefly, MDCK cells were cultured in UltraMDCK serum-free medium (Lonza) in the presence of 3 μM 15N-sphingolipid precursors (15N-sphingosine, and 15N-sphinganine) for 7 days. The medium was also supplemented with 50 μM 18O-cholesterol during the final 3 days. On day 6, cells were passaged onto poly-l-lysine coated pieces of silicon wafers and preserved the next day using glutaraldehyde and osmium tetroxide, as previously reported.1,141. H. A. Klitzing, P. K. Weber, and M. L. Kraft, Methods in Molecular Biology: Nanoimaging Methods and Protocols, edited by A. A. Sousa and M. J. Kruhlak (Humana, Totowa, NJ, 2013), Vol. 950, pp. 483–501.14. J. F. Frisz et al., Proc. Natl. Acad. Sci. U.S.A. 110, E613 (2013). https://doi.org/10.1073/pnas.1216585110 Cells on the silicon wafer substrate were imaged with low-dose scanning electron microscopy (JEOL 7000F scanning electron microscope, 1 keV) and coated with iridium. SIMS was performed on a Cameca NanoSIMS 50. The 12C14N−, 12C15N−, 16O−, 18O−, and 32S− secondary ions and secondary electrons were collected simultaneously. Depth profiling was performed of a 10 × 10 μm2 region using a 1.137-pA, 16-keV 133Cs+ primary ion beam. Images containing 512 × 512 pixels were collected.

B. Construction of 3D matrices

NanoSIMS data from 610 consecutive image planes from the depth profiling experiment were imported into the OpenMIMS plugin for imagej,38–4038. J. Schindelin et al., Nat. Methods 9, 676 (2012). https://doi.org/10.1038/nmeth.201939. P. Gormanns, S. Reckow, J. C. Poczatek, C. W. Turck, and C. Lechene, PLoS One 7, e30576 (2012). https://doi.org/10.1371/journal.pone.003057640. C. T. Rueden, J. Schindelin, M. C. Hiner, B. E. DeZonia, A. E. Walter, E. T. Arena, and K. W. Eliceiri, BMC Bioinf. 18, 529 (2017). https://doi.org/10.1186/s12859-017-1934-z and the pixel specific counts of 16O−, 18O−, 12C14N−, 12C15N−, and secondary electrons for each image plane were exported as .csv or .txt files. Using matlab (version R2020a and R2020b), the .csv or .txt files for each species were then compiled into separate 3D matrices containing the counts of that species at the x-, y-, and z-positions, where x and y are the pixel locations and z is the acquisition plane (Fig. 1, column 1). Planes 505, 506, and 521–530 did not contain usable data due to excessive charging. Therefore, planes 521–530 were removed from each 3D matrix, and planes 505 and 506 were replaced with planes 504 and 507, respectively.

C. Secondary ion matrix preprocessing

For noise reduction, the secondary ion counts at a given pixel were replaced with the average of the counts within a 5 × 5-pixel region centered on the given pixel. Then, the elements in the 16O− and 12C14N− matrices with 16O− or 12C14N− counts below 10% or 5%, respectively, of the maximum were set to zero to mask them. The counts of 18O− and 12C15N− at each position were ratioed to the counts of 16O− and 12C14N−, respectively, detected at the same position. The resulting ratios were subsequently divided by 0.002 005 or 0.003 67 to yield an enrichment factor that equals the number of times the abundance of 18O or 15N, respectively, is greater than terrestrial abundance.

D. Matrix alignment

Each z-plane in the secondary electron 3D matrix was aligned with the subsequent z-plane using an intensity-based registration function, imregister, in the matlab image processing toolbox (version 11.1). When significant lateral shifts prevented automatic alignment, we employed a custom manual alignment that allowed the user to select a visually distinct feature in each plane that was present in all the planes that required manual alignment. As an example, planes 500–540 were aligned using the circular feature visible in the yellow box near the left side of Fig. 5(i). After aligning the matrix of secondary electron image planes, the lateral translations for each plane were saved and applied to the 3D 15N- and 18O-enrichment matrices.

E. Morphology reconstruction from secondary electron images

To omit the substrate from the morphology reconstruction, the cell was manually outlined in the secondary electron image, and the portions of the image that were substrate and not cell were set to zero. To account for changes in the edge of the cell, a new mask was drawn every 50 image planes. To reduce the localized high secondary electron counts at each pixel that is caused by charging of the nonconductive biological sample, the angle of primary ion impact with the sample, and other factors unrelated to height, we surveyed three different illumination correction methods: logarithm transform,4141. C. C. Chude-Olisah, G. Sulong, U. A. K. Chude-Okonkwo, and S. Z. M. Hashim, in IEEE International Conference on Signal and Image Processing Applications (IEEE, Melaka, 2013), pp. 412–416. gamma correction, and matlab’s built-in contrast adjustment (imadjust). To apply the logarithm transform, the natural logarithm of the secondary electron pixel intensities in each matrix was calculated. Gamma correction was applied by raising each pixel intensity in the matrix to a value of 0.55. matlab’s imadjust function is used to scale the intensity values within a user-specified range to a new range. In this work, imadjust was used to reduce the values of the pixels with intensities in the top 50th percentile by a factor of one half. Then, the intensities of all the nonsubstrate pixels were decreased by a constant value to shift the baseline to zero.For cell morphology reconstruction, we estimated the height of the MDCK cell within the analysis region was 8 μm.42–4442. G. Barker and N. L. Simmons, Q. J. Exp. Physiol. 66, 61 (1981). https://doi.org/10.1113/expphysiol.1981.sp00252943. A. Jeanes, M. Smutny, J. M. Leerberg, and A. S. Yap, J. Mol. Histol. 40, 395 (2009). https://doi.org/10.1007/s10735-010-9253-y44. A. Puliafito, L. Hufnagel, P. Neveu, S. Streichan, A. Sigal, D. K. Fygenson, and B. I. Shraiman, Proc. Natl. Acad. Sci. U.S.A. 109, 739 (2012). https://doi.org/10.1073/pnas.1007809109 Based on the sputter rate determined for biological samples,4545. S. Ghosal, S. J. Fallon, T. J. Leighton, K. E. Wheeler, M. J. Kristo, I. D. Hutcheon, and P. K. Weber, Anal. Chem. 80, 5986 (2008). https://doi.org/10.1021/ac8006279 approximately 3.7 μm of material was sputtered from the cell before depth profiling began. Therefore, we estimated the cell material was 4.3 μm tall when the first depth profiling image was acquired. Because an additional 2.0 μm of material was sputtered from the cell during depth profiling (3.3 nm per image plane), we converted the pixels that corresponded to the cell in planes 2–600 to a height of 3.3 nm. Then, the pixel intensities in the first image plane were multiplied by a constant to convert the brightest pixel and thus tallest feature to a height of 2.3 μm so that the sum of planes 1–600 at each pixel location had a maximum height of 4.3 μm (Fig. 1, column 2). A topography matrix that contained the height at each x, y position on the surface of the cell at the beginning of the depth profile was saved.

The matlab function surf was used to create 3D surface renderings of the topography. To create an image of the sample at a desired image plane, n × 3.3 nm was subtracted from all nonzero pixels of the topography matrix, where n equals the number of the desired image plane.

F. Depth correction of the isotope enrichment matrices

The depth correction was applied to the isotope enrichment images by shifting the column of pixels with the same x, y coordinates the distance indicated by the value of the corresponding position in the topography matrix constructed in (Fig. 1, column 3). Applying this process to every x, y location of the 15N- and 18O-enrichment matrices produced depth corrected enrichment matrices in which the lateral position (x, y) of each cell was unchanged but the z-coordinate has shifted. Then, each cell in the 15N- and 18O-enrichment matrices was defined as a 19.5 × 19.5 × 3.3 nm3 voxel. The 15N- and 18O-enrichment matrices were visualized in 3D using the matlab function volview. A transparency of 85% was applied to the data for improved visualization.

III. RESULTS

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. EXPERIMENTIII. RESULTS <<IV. DISCUSSIONV. SUMMARY AND CONCLUSION...AUTHORS’ CONTRIBUTIONSREFERENCES

A. Morphology reconstruction

The previously reported data set1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681 used in this work was acquired from a 10 × 10 μm2 region on an MDCK cell that had been metabolically labeled so that it contained 15N-sphingolipids and 18O-cholesterol [Fig. 2(a)]. The analyzed region on the cell is indicated by the yellow box in Fig. 2(b). Based on the average sputter rate determined for other biological materials,4545. S. Ghosal, S. J. Fallon, T. J. Leighton, K. E. Wheeler, M. J. Kristo, I. D. Hutcheon, and P. K. Weber, Anal. Chem. 80, 5986 (2008). https://doi.org/10.1021/ac8006279 approximately 3.7 μm of material was removed from the analysis region prior to performing depth profiling. Another 2 μm of material was sputtered away during the analysis, leaving a small amount of cellular material on the substrate.1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681 Assuming a constant sputter rate, approximately 3.3 nm of material was removed from the cell each time the secondary ion beam was rastered across its surface. The secondary electron images acquired with the NanoSIMS were assembled into a stack such that the first image acquired was at the top of the stack and each subsequent image was placed below its predecessor. These images were aligned to correct for any drift in the sample or beam that may have occurred during acquisition and cropped so the same region was shown in every image in the stack. This produced a 3D image that showed a 7.8 × 8.8 × 2 μm3 region that was positioned approximately 3.7 μm below the cell surface. The same translations and cropping were applied to the 15N- and 18O-enrichment data that was acquired in parallel with the secondary electrons on the NanoSIMS.Our first task in reconstructing the cell morphology was to estimate the height of the tallest features shown in the first image obtained during depth profiling. We assumed the tallest features on the cell prior to NanoSIMS analysis were 8 μm based on literature reports of MDCK cells.42–4442. G. Barker and N. L. Simmons, Q. J. Exp. Physiol. 66, 61 (1981). https://doi.org/10.1113/expphysiol.1981.sp00252943. A. Jeanes, M. Smutny, J. M. Leerberg, and A. S. Yap, J. Mol. Histol. 40, 395 (2009). https://doi.org/10.1007/s10735-010-9253-y44. A. Puliafito, L. Hufnagel, P. Neveu, S. Streichan, A. Sigal, D. K. Fygenson, and B. I. Shraiman, Proc. Natl. Acad. Sci. U.S.A. 109, 739 (2012). https://doi.org/10.1073/pnas.1007809109 Because approximately 3.7 μm of material was sputtered from the analysis region prior to depth profiling, we estimated that the tallest features on the cell at the start of the depth profile were 4.3 μm. Therefore, we scaled the pixel intensities in the first image plane by multiplying by a constant such that the brightest pixel corresponded to a height of 2.3 μm, and the addition of the other 599 image planes that spanned a depth of 2 μm gave a total height of 4.3 μm. For the remaining image planes, the nonzero pixel intensities were replaced with 3.3 nm because the constant sputter rate assumption implies each layer has a uniform thickness. Finally, addition of the corresponding pixel locations in all the image plane matrices produced a single topography matrix that represents the cellular morphology (Fig. 1, column 2).Examination of the secondary electron image acquired for the first plane reveals multiple ridges on most of the upper left side of the cell and a smooth sloping surface on the right [Fig. 2(a)]. Some of the ridges on the cell toward the upper right corner of Fig. 2(a) appear bright due to charging and other effects that do not depend on the height of the feature such as the angle that the primary ion beam impacted the sample. This produced peaks that were too tall, giving the morphology a spiky appearance [Fig. 2(c)]. We surveyed various widely used illumination correction methods to eliminate these artifactual elevations in height.4141. C. C. Chude-Olisah, G. Sulong, U. A. K. Chude-Okonkwo, and S. Z. M. Hashim, in IEEE International Conference on Signal and Image Processing Applications (IEEE, Melaka, 2013), pp. 412–416. We began with the logarithm transform because it compresses bright pixels while enhancing dim pixels.46,4746. V. Dutt and J. F. Greenleaf, IEEE Trans. Med. Imaging 15, 802 (1996). https://doi.org/10.1109/42.54449847. A. V. Dsouza, H. Lin, J. R. Gunn, and B. W. Pogue, J. Biomed. Opt. 20, 1 (2015). https://doi.org/10.1117/1.JBO.20.8.080504 Although the logarithm transform compressed the high intensity pixels, it produced a surface that was too flat on the top and too tall at the edge of the cell [Figs. 2(e) and 2(f)]. We also tried a gamma correction of 0.55 because this correction adjusts the contrast in an image for optimal display on a monitor.4848. V. Damjanovski, CCTV (Elsevier, 2014), pp. 212–253. However, the dim pixel enhancement that resulted from this correction produced an inaccurate morphology at the cell edge (image not shown). To permit decreasing the highest intensity pixels without altering the lower values, we used matlab’s built-in contrast adjustment (imadjust) to decrease the intensities of the pixels within the top 50th percentile by a factor of one half. This produced peaks that were more consistent with visual perception of the secondary electron image [Figs. 2(g) and 2(h)] because it lowered the highest peaks on the cell without altering the morphologies of other features.

B. Depth correction of isotope enrichment data

The resulting topography matrix produced using the imadjust function was used to reshape the aligned and cropped 15N- and 18O-enrichment images that revealed the 3D distributions of 15N-sphingolipids and 18O-cholesterol in the MDCK cell. For comparison, recompiled versions of the previously published enrichment images without depth correction1717. A. N. Yeager, P. K. Weber, and M. L. Kraft, Biointerphases 11, 02A309 (2016). https://doi.org/10.1116/1.4939681 are shown in Figs. 3(a) and 3(b) and 1(a) and 1(b) in the supplementary material.5757. See the supplementary material at https://www.scitation.org/doi/suppl/10.1116/6.0001092 for a supplemental figure showing the top view of the uncorrected and depth corrected isotope enrichment 3D images of isotope-labeled sphingolipid and cholesterol distributions within a depth profiled region of a cell. We created a more accurate rendering of subcellular morphology by shifting each 19.5 × 19.5 × 3.3 nm3 voxel to the depth corrected z-position that was specified by the topography matrix. We assessed whether these new z-positions moved a significant percentage of the voxels below the substrate defined by z = 0, which would indicate the cell height had been underestimated. Only 0.11% and 0.09% of the voxels in the depth corrected 15N- and 18O-enrichment images, respectively, were below the z = 0 μm boundary. These percentages verify the height of the MDCK cell within the analysis region at the start of depth profiling was not underestimated. Hence, this also suggests that our approach may be used to depth correct other 3D SIMS images of cells as long as reasonable estimations of sputter rate and total cell height are available.The resulting depth corrected enrichment images [Figs. 3(c) and 3(d) and 1(c) and 1(d) in the supplementary material]5757. See the supplementary material at https://www.scitation.org/doi/suppl/10.1116/6.0001092 for a supplemental figure showing the top view of the uncorrected and depth corrected isotope enrichment 3D images of isotope-labeled sphingolipid and cholesterol distributions within a depth profiled region of a cell. conform to the contours of the cell. Due to the curved cell surface, the image shows a 2 μm thick ribbon of cellular material that is bound within a 7.8 × 8.8 × 4.3 μm3 region [Figs. 4(d)–4(f)]. Unlike the uncorrected 3D 18O-enrichment image [Fig. 3(b)] that shows a cholesterol-rich strip at the top of the image, the depth corrected image [Fig. 3(d)] accurately shows this thin cholesterol-rich strip near the substrate. Because this depth profile was collected from a region 3.7 μm below the cell surface, the plasma membrane at the top of the cell was removed prior to the analysis. Therefore, the plasma membrane should not be present at the top of the image, and the cholesterol-rich strip corresponds to a small portion of the basolateral plasma membrane beneath the cell. The basolateral plasma membrane is only visible at the edge of the cell because this is the only location where the cell’s thickness was less than the depth of the material sputtered from the sample with SIMS. The even cholesterol distribution within this strip is consistent with the previous observation that cholesterol is evenly distributed in plasma membranes.12–1412. R. L. Wilson, J. F. Frisz, H. A. Klitzing, J. Zimmerberg, P. K. Weber, and M. L. Kraft, Biophys. J. 108, 1652 (2015). https://doi.org/10.1016/j.bpj.2015.02.02613. J. F. Frisz, H. A. Klitzing, K. Lou, I. D. Hutcheon, P. K. Weber, J. Zimmerberg, and M. L. Kraft, J. Biol. Chem. 288, 16855 (2013). https://doi.org/10.1074/jbc.M113.47320714. J. F. Frisz et al., Proc. Natl. Acad. Sci. U.S.A. 110, E613 (2013). https://doi.org/10.1073/pnas.1216585110 Additionally, the cholesterol-enriched tubules near the right side of the uncorrected image [Fig. 3(b)] are spherical in the corrected image [Fig. 3(d)]. Though not apparent in the uncorrected image due to the position of the cholesterol-rich strip that represents a portion of the plasma membrane, the spherical cholesterol-rich features only span approximately 75% of the analysis depth [Fig. 1 in the supplementary material].5757. See the supplementary material at https://www.scitation.org/doi/suppl/10.1116/6.0001092 for a supplemental figure showing the top view of the uncorrected and depth corrected isotope enrichment 3D images of isotope-labeled sphingolipid and cholesterol distributions within a depth profiled region of a cell. The voxels with low 18O-cholesterol enrichment above and below the cholesterol-rich features are more easily seen in the depth corrected image because those at the top are no longer covered by the cholesterol-rich strip, and those beneath it are more easily seen when the data is on the 4.3 μm z-scale that is needed to show the cell’s curvature in the depth corrected image.The uncorrected 3D 15N-enrichment image in Fig. 3(a) shows sphingolipid clusters that appear to be randomly dispersed. In contrast, in the depth corrected 15N-enrichment image [Fig. 3(c)], small sphingolipid-rich regions are dispersed throughout the volume but the sphingolipids appear to form spherical objects that seem to be located near the cholesterol-rich spheres. Cross sections taken across the largest cholesterol-rich volume in the 15N- and 18O-enrichment images and overlays reveal the 15N-sphingolipid enrichment surrounds the 18O-cholesterol-rich spherical objects [Fig. 4]. The horizontal cross section taken at z = 2.1 μm [Figs. 4(d)–4(f)] bisects both cholesterol-enriched spheres that are visible in Fig. 4(d). The perimeters of these structures have 15N-enrichments that vary from 3 to 5 times that of the natural abundance for nitrogen and 18O-abundances that are 3 to 20 times the natural abundance level of oxygen. This suggests the cholesterol-rich features are enclosed in membranes that contain sphingolipids and cholesterol. The enriched features have a horizontal diameter between 1.4 and 1.8 μm [Figs. 4(a) and 4(b)].

C. Evaluation of reconstructed morphology

We evaluated the extent that the features in our reconstruction matched those shown in the secondary electron images acquired at increasing depths. In Fig. 5, the secondary electron images that were acquired with the NanoSIMS instrument at the first, 100th, 200th, 300th, and 600th image plane are shown in the first column and the corresponding reconstructed morphologies are shown in the second column. Based on comparison to the first secondary electron image [Fig. 5(a)], the corresponding reconstruction [Fig. 5(b)] accurately captures the basic shape of the cell as well as its fine features. The numerous ridges on top of the cell that can be seen in the secondary electron image are reproduced in the reconstruction. As an example, two parallel ridges that run vertically are present near the upper right corners of both the secondary electron and reconstruction images. However, some of the features along the cell’s edge are not resolved in this reconstruction. This is due to our substrate masking procedure, which uses the same mask to define the edge of the cell for 50 consecutive image planes. Future work will address this issue. Nonetheless, the features in our reconstruction of the surface of the cell at the start of depth profiling match the ridges that are visible on the cell in the first secondary electron image. Therefore, a good approximation of the cell morphology may be constructed by using the intensities of the pixels in the secondary electron images acquired during depth profiling SIMS.Comparison of the secondary electron image acquired at the 100th image plane to the corresponding reconstruction yielded similar conclusions to those drawn for the first plane. This was also largely true for the 200th image plane and the corresponding reconstruction. However, discrepancies between the cell surface features shown in the secondary electron images and reconstruction become evident at image plane 300 [Figs. 5(g) and 5(h)]. The secondary electron image shows a small circular depression toward the left edge of the analysis region that is missing in the corresponding reconstruction. This region is outlined by the left yellow box in Figs. 5(g) and 5(h). Instead of a circular depression, a few small ridges are present at this site in the reconstruction. As sputtering continued, the size of the circular depression increased, as shown in the 600th image plane [Fig. 5(i)], but this feature was still absent from the corresponding reconstruction [Fig. 5(j)]. A second less noticeable discrepancy is the u-shaped ridge near the top right of the analysis region in the secondary electron image acquired at plane 300 has a slightly different shape and location in the corresponding reconstruction [right yellow boxes in Figs. 5(g) and 5(h)]. This discrepancy is more evident for the 600th image plane [Figs. 5(i) and 5(j)].

Because the reconstructions well approximated the cell morphologies shown in the secondary electron images at the beginning of the depth profile but not after hundreds of image planes were acquired, we attribute the discrepancies observed after plane 200 to a nonconstant sputter rate. Therefore, we conclude that our reconstruction of cell morphology based on secondary electron pixel intensity is accurate, but our assumption of a constant sputter rate is not.

IV. DISCUSSION

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. EXPERIMENTIII. RESULTSIV. DISCUSSION <<V. SUMMARY AND CONCLUSION...AUTHORS’ CONTRIBUTIONSREFERENCES3D SIMS images may be produced by sequentially stacking the 2D SIMS images acquired during depth profiling. However, when 3D SIMS images are created of nonplanar samples, the depths of the secondary ions do not reflect the original positions of the parent molecules in the intact sample. This is because depth in the 3D SIMS