2025 Theses Doctoral

Human brain optimized light sheet microscopy (HOLiS) for high-throughput cell type atlasing of whole human brains

Casper, Malte Johannes

In a human brain around 200 billion cells are working together shaping who we are. Grasping their complex organization, even in a single brain remains a major challenge. Advances in tissue clearing provide a window onto cells inside of intact brains, and have recently been optimized for clearing and immunostaining human brain tissue. However, imaging cleared tissues on the scale of the whole human brain presents many challenges, with conventional methods likely to require many months or years of acquisition time per brain.

The work of this thesis presents a complete imaging and analysis pipeline developed specifically to achieve high-throughput, multispectral imaging of entire, cleared and immunolabelled human brains at cellular resolution for cell type atlasing with acquisition times of less than 2 weeks. Our human brain optimized light-sheet (HOLiS) microscopy system is a form of oblique-plane single objective light-sheet, capable of imaging 5 mm thick, optically cleared, complete coronal sections of human brain.

By imaging thick sections, we can reduce tissue deformations and cut-edge effects, although this necessitated novel optical solutions to enable implementation of a long working distance, multi-immersion primary objective lens. Another feature of HOLiS is its ability to image multiple spectral channels in parallel to enable multiplexed antibody labeling of the different cell types. Combining multiple laser lines for simultaneous excitation, emitted fluorescence is spectrally divided by our novel 4-way image splitter for simultaneous detection of 5 spectral channels with the capacity for 9 channels imaged in parallel. The 5th channel images a nuclear dye, providing fiducials for every cell. Although spectral multiplexing adds complexity, it greatly accelerates HOLiS acquisition time and reduces data processing burden. Using ultra-fast cameras, we acquire multi-spectral images up to at 0.75 mm³/sec with micron sampling translating to image an entire human brain within 2 weeks.

At modest sampling density and 9 spectral channels, the data of a single human brain scan is expected to exceed 3 PB, imposing constraints on transfer, storage, data pre-processing, and accessibility. Using a parallelizable analysis pipeline which locates every nucleus, and extracts cell type information from the spectral channels to provide compressed point-cloud representations of the data that can be quantitatively analyzed, shared and compared between brains. These rich datasets can be clustered to explore cell type distributions and used to guide more complex feature-based analysis, neuroanatomical segmentation and efficient visualization of raw data.