Carl Zeiss (United States)

Proteins can diffuse micrometers in seconds, yet bacterial cells are able to maintain stable protein gradients. The best studied bacterial protein gradient is the Min system of Escherichia coli. In rod-shaped bacteria the MinCD proteins prevent formation of minicells by inhibiting FtsZ polymerization close to the cell poles. In E. coli these proteins oscillate between cell poles within a minute, resulting in an increased MinCD concentration at the poles. This oscillation is caused by the interaction between MinD and the protein MinE, which form an ATP-driven reaction-diffusion system, whereby the ATPase MinD cycles between a monomeric cytosolic and a dimeric membrane attached states. Bacillus subtilis also has MinCD, but lacks MinE. In this case MinCD form a static gradient that requires the transmembrane protein MinJ, located at cell poles and cell division sites. A recent reaction-diffusion model was successful in recreating the MinD gradient in B. subtilis, assuming that MinD cycles between cytosol and membrane, like in E. coli. Here we show that the monomeric and dimeric states of B. subtilis MinD have comparable membrane affinities, that MinD interacts with MinJ as a dimer, and that MinJ is not required for membrane localization of MinD. Based on these new findings we tested different models, using kinetic Monte Carlo simulations, and found that a difference in diffusion rate between the monomer and dimer, rather than a difference in membrane affinity, is important for B. subtilis MinCD gradient formation.

Abstract Early detection of retinal molecular biomarkers is crucial for addressing the unmet clinical need to prevent irreversible neural tissue damage in ophthalmic and neurodegenerative diseases. Among emerging molecular sensing techniques, non-resonant Raman spectroscopy stands out as a naturally label-free and noninvasive method, offering rich biochemical information. However, in vivo detection of non-resonant Raman spectra from retinal tissue has proven to be challenging so far. Previous studies have reported conflicting results, likely due to overwhelming pigment autofluorescence. In this study, we identified the optic nerve head as the optimal retinal location for acquiring non-resonant Raman spectra in the molecular fingerprint region. Through longitudinal intra-subject measurements, we revealed dynamic changes in the molecular composition. Furthermore, a comparative study across age groups enabled the identification of molecular alterations associated with aging. These findings establish a critical foundation for utilizing non-resonant Raman spectroscopy as an early diagnostic tool for the detection of molecular biomarkers associated with ophthalmic and neurodegenerative diseases.

Non-destructive 3D X-ray microscopy (XRM) is widely adopted in package-level failure isolation and analysis [1]–[2]. However, the imaging resolution is generally limited to submicron due to optical diffraction and penumbra blurring of X-ray source spot. As emerging packaging techniques such as <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2.5 \mathrm{D} / 3 \mathrm{D}$</tex> heterogeneous integration increasingly involve small-volume bumps and hybrid bonds, resolution breakthrough becomes indispensable for effectively analyzing these downsized interconnects. 3D X-ray nanotomography technique, known as “nanoCT”, is capable of imaging semiconductor packages and interconnects with fifty-nanometer resolution [3]–[4]. However, long data acquisition time has limited its practical application, despite its superior resolution over most non-destructive methods. In this report, we will demonstrate how a deep learning highresolution reconstruction workflow (referred as “DeepRecon” herein) through trained neural networks, can improve scan speed by a factor of four. Several semiconductor package samples with hybrid bonds will be analyzed using nanotomography and the proposed reconstruction method. With the improved scan speed by AI, a correlated FIB-SEM workflow is followed to further analyze fault regions with nanometer accuracy.

Abstract Image segmentation is a valuable tool for visual image data inspection of semiconductor device structures. For the large amounts of data provided by recent advancements in automated scanning electron microscope (SEM) and focused ion beam-scanning electron microscope (FIB-SEM) data acquisition, automatic segmentation becomes indispensable to fully exploit the information contained in the data in automated characterization workflows. Using two exemplary FIB-SEM tomography datasets, we explored artificial intelligence-based image segmentation using only a minimum amount of training images annotated by a human user. Furthermore, we explored possibilities to automatically localize defects based on the segmentations.

Innovations in 3D tissue imaging have revolutionized research, but limitations stemming from lengthy protocols and equipment accessibility persist. Classical widefield microscopy is fast and accessible but often excluded from 3D imaging workflows due to its lack of optical sectioning. Here we combine tissue clearing with a depth-variant deconvolution approach customized for large-volume widefield imaging to achieve subnuclear axial resolution in tissues to a depth of 500 μm. We illustrate the utility of this method in a mouse model of ileitis and to gain a 3D perspective in thick brain slices from a mouse model of cerebral amyloid angiopathy, where we resolved large and small blood vessels, including those with amyloid deposits, attaining resolution that compared favorably to tile-scanning confocal microscopy. Finally, we sought to leverage our approach to allow for richer pathological evaluation of human kidney biopsies. Our approach produced hundreds of consecutive z-planes in five minutes of imaging for 3D visualization of winding arterioles feeding into glomeruli. This 3D perspective afforded straightforward identification of atrophic tubes in fresh kidney biopsies prepared in 2 hours to simulate the time-constrained evaluation of donor kidneys for transplant suitability. Having achieved subnuclear z-resolution in sections hundreds of microns thick, widefield microscopy coupled to robust deconvolution now emerges as an accessible and viable method to gain 3D insight in research or clinical pathological evaluations.

In vivo access to molecular information of retinal tissue is considered to play a critical role in enabling early diagnosis of ophthalmic and neurodegenerative diseases. The current gold standard of retina imaging, optical coherence tomography and angiography provides only the retinal morphology and blood perfusion, missing the full spectrum of molecular information. Raman spectroscopy addresses this gap while keeping the investigation non-invasive and label-free. Although previous studies have demonstrated the huge diagnostic potential of combining both modalities for in vivo biological tissue measurement, some have either employed unsafe optical power levels for in vivo retinal measurements or presented results that were negative or contradictory. In this study, we have developed an eye-safe multimodal in vivo label-free imaging system and demonstrate the potential of this device by investigating the retina of a living albino rat. The acquired Raman spectra showed relevant Raman bands in comparison with the previous ex vivo studies. Using this multimodal imaging system for non-invasive retina measurements of transgenic rodents holds the potential to advance the understanding of the pathophysiology of both ophthalmic and neurodegenerative diseases.