Abstract
Recently, high-throughput image-based transcriptomic methods were developed and enabled researchers to spatially resolve gene expression variation at the molecular level for the first time. In this work, we develop a general analysis tool to quantitatively study the spatial correlations of gene expression in fixed tissue sections. As an illustration, we analyze the spatial distribution of single mRNA molecules measured by in situ sequencing on human fetal pancreas at three developmental time points—80, 87, and 117 days post-fertilization. We develop a density profile-based method to capture the spatial relationship between gene expression and other morphological features of the tissue sample such as the position of nuclei and endocrine cells of the pancreas. In addition, we build a statistical model to characterize correlations in the spatial distribution of the expression level among different genes. This model enables us to infer the inhibitory and clustering effects throughout different time points. Our analysis framework is applicable to a wide variety of spatially-resolved transcriptomic data to derive biological insights.
Introduction
The field of transcriptomics has moved beyond simply identifying which genes are expressed to determining exactly where they are located within a tissue. Technologies like in situ sequencing (ISS) provide high-resolution maps of individual mRNA molecules, yet the computational tools to analyze this spatial "social behavior" have lagged. Most studies still rely on qualitative observations. This paper introduces a rigorous statistical framework to quantify how genes cluster, avoid each other, or relate to physical structures like cell nuclei and organ-specific features.
Quantitative Spatial Analysis Framework
The researchers developed a two-pronged computational approach to move beyond visual inspection:
- Density Profile-Based Analysis: This method measures the density of gene transcripts as a function of their distance from morphological landmarks, such as the boundary of a pancreatic islet or a cell nucleus. By using Kernel Density Estimation (KDE) to smooth the data and Kullback-Leibler (KL) divergence to compare these "spatial signatures," the team could statistically group genes with similar distributions.
- Multitype Strauss Process Model: To understand direct gene-gene interactions, the authors adapted a spatial point process model. Unlike simpler methods, this model distinguishes between genes that co-occur by chance and those with a functional "pull" (clustering) or "push" (inhibition). It calculates a correlation parameter ($\gamma_{ij}$), where values $>1$ indicate clustering and values $<1$ indicate inhibition.
Insights from the Developing Human Pancreas
The study applied these tools to a longitudinal dataset of the human fetal pancreas at 80, 87, and 117 days post-fertilization (dpf). At the 117-day mark, the dataset included over 159,000 detected RNA molecules across roughly 50,000 cells.
Spatial Maturation and the MUC6 Discovery
The analysis revealed that as the fetus develops, the spatial separation between functional groups of genes becomes significantly more pronounced. While patterns are diffuse at day 80, the molecular architecture of the islets is sharply defined by day 117.
A key discovery was the unexpected spatial behavior of MUC6. Traditionally known for mucus production in epithelial tissues, MUC6 transcripts were found to progressively "migrate" in their expression pattern to cluster with known endocrine markers like insulin (INS) and glucagon (GLUC). This suggests MUC6 may play a previously unrecognized role in the differentiation of pancreatic progenitor cells.
Quantitative Interaction Mapping
The KL divergence analysis (Figure 6) successfully clustered the 25 targeted genes into two distinct spatial groups. Group 1—including INS, GLUC, SST, PDX1, EPCAM, PROM1, and MUC6—showed high density near or within islets.
The Multitype Strauss model further refined these insights. It showed that while endocrine markers like SST and INS are found in the same regions, they have correlation values near 1.0 at short ranges (20μm), reflecting that they are expressed in distinct, neighboring cells. Conversely, the highest clustering effects were seen between EPCAM ↔ PROM1 ($\gamma = 1.33$) and MUC6 ↔ PROM1 ($\gamma = 1.19$), suggesting these transcripts are often physically co-located within the same progenitor cells.
Limitations and Future Outlook
The current study is limited by a relatively small gene panel (25 genes) and the inherent challenges of automated cell segmentation in complex tissues. Additionally, ISS has a lower detection efficiency compared to methods like smFISH. However, this framework is "technology-agnostic" and can be applied to any spatial transcriptomic data, such as the Human Cell Atlas, to identify new diagnostic targets and compare healthy versus diseased tissue architectures.
Summary updated