HuBMAP Groups Come at Human Kidney Anatomy from Bottom-Up, Top-Down

HuBMAP Groups Come at Human Kidney Anatomy from Bottom-Up, Top-Down

February 8, 2021

This HuBMAP SciTech Webinar featured two very different projects that come at the question of the anatomy and function of the human body from opposite directions. For the first time, researchers are able to detect lipid molecules in microscopic locations, and have developed a new framework to align human kidney images automatically. By combining these approaches, researchers can approach a common molecular and anatomical understanding of kidney function from both the bottom-up and the top-down.

Multimodal nano-DESI and MALDI IMS for Lipid Investigation of the Human Kidney

Elizabeth Neumann of Vanderbilt University and Julia Laskin of Purdue University described how their collaboration, led by Jeff Spraggins of Vanderbilt, is using two different types of imaging mass spectrometry to investigate the localization of lipids—an important but under-studied type of biomolecule—in the human kidney. Imaging mass spectrometry enables the detection of endogenous biomolecules without the need for any chemical tags. The experiment is performed by desorbing and ionizing molecules at discrete locations on the surface of a tissue and detecting them with a mass spectrometer. The intensity of each biomolecule is then plotted at each location, or pixel, to create a molecular image. Since biomolecules have a more or less unique molecular weight, and mass spectrometry can detect tiny differences in mass, scientists can identify each substance within a sample.

The team is using MALDI and the newer nano-DESI, two different means of imaging tissue samples with mass spectrometry at microscopic scales, in concert. MALDI works by first coating the tissue surface with a small molecule, typically called a matrix, that helps with the extraction and ionization processes. A laser is then focused to a very small point on the tissue causing biomolecules in the tissue to be desorbed and ionized so they can be detected by the mass spectrometer. In nano-DESI, on the other hand, the sample doesn’t require any matrix and can be analyzed “naked.” Using a specially-designed microfluidic probe, a very small solvent droplet can be formed at the tissue surface that allows scientists tolift biomolecules from that point and inject them into the mass spectrometer. Using MALDI and nano-DESI in consecutive slices from a kidney tissue only 10 micrometers thick (less than one ten-thousandths of an inch) and in similar areas of each sample, the researchers compared the signals of the two methods.

One big takeaway of the work to date is that both methods can identifyand image lipids—the fatty molecules that serve as structural components in cellular membranes, a source of energy, and signals in some cellular processes—with good sensitivity and spatial resolution. The other finding is that MALDI and nano-DESI result in overlapping but distinct signals—they identify roughly 50 percent of the same molecules, with each identifying a broad range of molecules the other can’t. The partially overlapping signals offer a new way of identifying and classifying how lipids and other molecules differ in health and disease, with early work showing distinct mixtures of lipids in different parts of the kidney and in healthy versus abscessed kidney tissue. While the work is a fundamental attempt to identify and understand the locations and roles of these molecules, the vast number of new molecules it identifies offers the possibility of obtaining an exquisitely fine “molecular fingerprint” for identifying normal and disease processes in the kidney and diagnosing kidney disease sooner, when it may be easier to treat.

Design and Construction of 3-D Atlases to Support Multi-Scale Anatomical Mapping

Bennett Landman of Vanderbilt presented a very different project, which is studying kidney structure from the top down, beginning at the level of the whole organ. A major problem in studying human kidney images is that the size and shape of healthy human kidneys varies a lot between individuals. People with no evidence of kidney disease in their clinical records can have kidneys that vary in volume by 300 times, and the organs’ positions in the body can vary enough that identifying images on a CT or MRI scan requires a trained radiologist or other doctor. Being able to identify and align kidney images automatically could help scientists study hundreds or thousands of the organs at once, identifying differences that could predict future kidney disease.

The team’s task was made more complicated by the fact that there are different types of kidney scans that give different-looking pictures of the same kidney. CT scan contrasts rely on an X-ray-opaque dye flushing through a patient’s kidney’s blood vessels, and so taking scans at different times after injection show different parts of the organ—and individual’s kidneys flush the dye at different rates. In addition, multiple distinct types of MRI scans complicate the task of automated identification and alignment of kidney images.

The scientists’ project to date has accomplished two major milestones. One is that they have created a coordinate system that can align images of the organ in an apples-to-apples way. Their system can now identify and compare different types of image—whether from CT to MRI and back or between types of each—and align them so that clear anatomical comparisons can be made at the speed of light.

Another milestone was making the system able to handle partial images—since for a particular patient doctors may only scan a part of the body—as well as accurate and fast identification of what part of the body a given image represents. The scientists also aimed to overcome occasional missing or mislabeled information in the scans, and to tackle the tricky problem of how individual patients’ metabolisms alter how the images appear. Initial results using “deep learning” artificial intelligence to train a computer to pick out details beyond the ability of the expert human eye to distinguish have been encouraging, with accuracies above 90 percent.

Future goals include using deep learning to identify and align tiny substructures in the kidney and to expand the work to the pancreas, spleen and eye. The researchers would also like to study the variation in “normal” kidneys more intensively, to see if the extremes may be signs of future kidney disease before any symptoms appear.