.

Related articles

Is healthy bone compatible with the Visium CytAssist Spatial Gene Expression for FFPE assay?

Question: Is healthy bone compatible with the Visium CytAssist Spatial Gene Expression for FFPE assay?

Answer: Depending on the species and type of bone, it may be compatible with our standard (described in Demonstrated Protocols and User Guide) Visium CytAssist Spatial Gene Expression for FFPE assay protocols and generate acceptable data. The challenges of using bone samples with on-slide biochemical assays stem from tissue detachment and the tissue typically being RNA-poor. This can result in data generated that lacks spatiality, complexity, and sensitivity as illustrated in key metrics reported in the Web Summary File (HTML output from Space Ranger). 

From limited testing, we have found that mouse bone showed expected spatiality while following 10x Genomics protocols but human bone had very poor spatiality, complexity, and sensitivity even with protocol modifications. Because data quality from bone samples may be variable, it is listed as a ”challenging tissue” type in our Tested Tissue List. We would recommend performing a small pilot experiment with bone tissue before scaling up. 

Below is a summary of our learnings and considerations from working with healthy mouse femur and tibia bone FFPE tissue sections. We have minimal recommendations for working with human bones based on experimental data generated from femur. Bone samples outside of those noted above have not been tested. Modifications to documentation (Demonstrated Protocols and User Guide) steps are unsupported. We cannot guarantee assay performance even when implementing the guidance listed in this article and therefore, customers can proceed at their own risk. 10x Genomics Support does not have additional information beyond what is provided in this article regarding processing bone FFPE tissue sections and will not be able to provide additional details from internal experiments or guidance. 

Fixation:

  • Tissue should be fixed with fresh formaldehyde solution 
  • For larger samples, generate a coronal or axial section with a saw, scalpel, or scissors in order to achieve more consistent fixation
  • Optimal fixation time will be variable depending on the composition and size of the sample
  • General recommendations and best practices can be found in this Leica article: An Introduction to Specimen Processing 

Decalcification:

Decalcification is required to perform histological, immunological, and molecular studies on bone tissue but a consequence of decalcification is RNA fragmentation. Thus, optimizing conditions that maximize bone softening while retaining RNA integrity is critical for performing molecular experiments with bone. Decalcification can be performed on the surface of an FFPE block or it can be performed during or after fixation, prior to embedding. In these studies, we tested decalcification immediately after fixation and thus, other decalcification methods will require additional optimization. As EDTA-based decalcification methods preserved nucleic acid quality over acid decalcification methods (e.g., formic, hydrochloric, and nitric acids)[1,2], acid decalcification methods were not tested internally. We exclusively utilized 0.5 M EDTA pH 8 for decalcification during internal experiments. Below are recommendations based upon limited testing.

  • EDTA powder will not dissolve in more acidic solutions, ensure that the solution is at pH 8 so EDTA is fully in solution
  • Utilize nuclease-free reagents so as to not introduce nucleases
  • Ensure to remove as much muscle and soft tissue as possible as it will impinge on the decalcification process  
  • Fully submerge tissue in a sufficient volume of EDTA solution and utilize a stir bar or nutator to gently agitate the solution during decalcification  
  • EDTA solution should be refreshed periodically (e.g., every 1-2 days if optimal fixation time is 1-2 weeks) by removing old solution and adding new solution to enhance decalcification process 
  • Monitor progress periodically (i.e., daily or weekly depending on size of sample) 
  • Decalcification is complete once the bone is pliable (as determined by gently poking the tissue with sharp metal forceps)
    • Larger bones (like mouse tibia) may require weeks while smaller bones (like biopsy specimens) may require only a few days for optimal decalcification
    • Human bone may take months for optimal decalcification depending on the size and bone region
  • Wash the decalcified bone tissue several times with 1x PBS pH 7.4, then proceed to processing and embedding 

Processing and Embedding:

  • To minimize RNA fragmentation, use low-melting paraffin wax for embedding
  • General recommendations and best practices can be found in this Leica article: An Introduction to Specimen Processing 

Block Selection: 

  • We recommend that RNA Quality Assessment is performed as described in the Tissue Preparation Guide
    • Be mindful that surrounding muscle and soft tissue may artificially inflate the DV200 values of the bone sample, impacting accuracy of RNA quality assessment of the bone itself
    • It is important to note that bone samples are often RNA-poor so it is critical that the RNA concentration is determined and the appropriate automated electrophoresis instrument, reagents, and chip are selected
    • We do not have a minimum RNA yield threshold to perform the Visium assay, as correlations between bulk RNA yield and assay performance have not been made
  • During sectioning, if there is evidence of insufficient decalcification, it may be necessary to select a different block (see below for additional information)

Sectioning:

  • Ensure to clean the workstation, microtome, tools, and blades with RNase Decontamination Solution before sectioning to minimize introducing exogenous RNases
  • Utilize a clean, new blade for sectioning 
  • 3-10 µm section thickness has been validated with the Visium CytAssist Spatial Gene Expression for FFPE assay but utilizing thinner bone sections (3-5 µm) may help promote tissue adhesion during the workflow
    • During testing, we utilized 5 µm sections with bone samples
  • If the blade is chattering during sectioning or the block is too hard to generate a good quality section, the block may not be efficiently decalcified
    • Consider warming the block using lukewarm water (37°C) to soften and hydrate it prior to sectioning
    • Similar protocols suggest performing surface decalcification and resectioning, but this method has not been validated with this assay
  • During limited testing with healthy bone, healthy skin, and human breast, we found that sample adhesion was significantly improved when sections were placed on specially coated hydrogel slides 

Tissue Staining:

User Guide: 

  • Given the lower RNA content of bone, the signal-to-noise ratio may be poor, leading to lower usable reads, which may be improved by adjusting steps in the User Guide
    • We found that performance of mouse and human bone did not improve with using Direct Placement (Visium Spatial Gene Expression for FFPE v1) Pre-Hybridization conditions (see: Visium Spatial Gene Expression Reagent Kits User Guide)
    • We have not tested adjusting the Probe Release conditions with either mouse or human bone samples 

Mouse Femur:

Below are observations made from a representative Web Summary File when running mouse femur on Nexterion Hydrogel tissue slides without any additional protocol modifications.

  • The Visium library generated by this experiment was good quality, as Valid Barcodes and Valid UMIs were high (>99%). Furthermore, given that the Q30 scores were high (>93%), sequencing performed as expected. 
  • In the “Mapping” section, the Reads Mapped to Probe Set metric is high (~98%), in addition to the Reads Mapped Confidently to Probe Set (~96%). Some usable reads were lost from the Reads Mapped Confidently to the Filtered Probe Set metric (~87%), indicating that the majority of reads that were confidently mapped were not filtered out due to predicted off-target activity. Together, this indicates that RNA templated ligation (RTL) performed as expected. 
    • See this Q&A article for details on mapping. 
  • In the last section of the “Summary” tab, the “Spots” metrics indicate that the majority of library molecules map back to tissue covered spots (Fraction Reads in Spots Under Tissue; ~98%). While the total number of genes detected was high (~19,000 of the total 20,551 genes), the Median UMIs per Spot and Median Genes per Spot were lower than expected (<2,000 for both). 
  • Within the “UMIs Detected” portion of the “Gene Expression” tab we see that the UMIs across the majority of the tissue are very low. The regions of the tissue with high UMIs (>2,000 UMIs), were primarily localized to the bone marrow and muscle (Figure 1A & B). Together, this indicates that the majority of UMIs originate from the muscle and bone marrow of the sample and minimal signal from the bone itself. 
  • Based on the “Clustering” section of the Web Summary File, the sample had clusters that aligned well with the general tissue morphology (Figure 1C). The associated t-SNE Projection illustrates that the clustering is distinct and organized and that spatiality is likely maintained fairly well (Figure 1D). 

 

Figure 1. Key Web Summary File plots generated from mouse femur tissue. (A) Tissue image with (B) UMIs and clustering (C) overlaid or (D) projected. 

  • From the Top Features by Cluster table, there were several differentially expressed genes and the top genes agreed with the tissue morphology. For instance, in Cluster 1 (blue; Figure 1C) the top features (Hbb-bt, Bpgm, and CD79a) are canonically expressed in bone marrow and agree with the bone marrow identified in the tissue image. Contrarily, markers of cartilage and bone were fairly lowly expressed in Cluster 2 (Col2a1 and Bglap, respectively) despite cartilage and bone being identified in the tissue image. Lastly, Myh2 (a muscle marker) was found to be expressed in Cluster 4 which agreed with the muscle identified in the tissue image. 

Across the mouse samples tested, we were able to generate Visium CytAssist Spatial Gene Expression libraries with good quality spatiality. The complexity and sensitivity of the mouse bone samples are lower in the bone and cartilage regions but well-maintained in the bone marrow. Thus, improvements in sensitivity in bone and cartilage may be improved by modifying the workflow.

Figure 2. Plots showing canonical markers overlaid on mouse femur tissue. (A) Bglap is expressed in bone. (B) Bpgm is expressed in red blood cells in bone marrow. (C) Col2a1 is expressed in cartilage. (D) Myh2 is expressed in muscle. 

Human Bone:

Below are observations made from a representative Web Summary File when running human femur on Nexterion Hydrogel tissue slides without any additional protocol modifications.

  • The Visium library was of good quality as Valid Barcodes and Valid UMIs were high (>98%). Furthermore, given that the Q30 scores were high (>95%), sequencing performed as expected. As the “Sequencing” metrics were typical, the poor performance was likely due to the sample itself. 
  • In the “Mapping” section, the Reads Mapped to Probe Set metric is high (>95%) but the Reads Mapped Confidently to Probe Set was very poor (~55%). High chimeric reads can typically originate due to poor RNA quality, Post-Hybridization/Ligation Washes, or detachment. When reviewing the data, most reads lost were due to chimeras while the rest were primarily singlet probes. Minimal usable reads were lost from the Reads Mapped Confidently to the Filtered Probe Set metric (~53%), indicating that the majority of reads that were confidently mapped were not filtered out due to predicted off-target activity. As there was no detachment or deviation in the washing protocol, the likely root cause of the low usable reads is low yield, poor quality RNA leading to very little on-target RNA templated ligation (RTL) product. 
    • See this Q&A article for details on mapping. 
  • The “Spots” metrics indicate that the majority of library molecules map back to tissue covered spots (Fraction Reads in Spots Under Tissue; >80%). And while the total number of genes detected was high (~18,000 of the total 18,536 genes), the Median UMIs per Spot and Median Genes per Spot were very low (<150 for both). Together, this indicates that the tissue sample had very poor sensitivity across the tissue section. 
  • Within the “UMIs Detected” portion of the “Gene Expression” tab we see that the UMIs across the majority of the tissue are very low. The regions of the tissue with high UMIs (~1,000; red color), were fairly sparse. Portions of the tissue that were more cell-dense (Figure 3A), had very low UMI counts (<500 UMIs; Figure 3B). Taken together, this indicates that the majority of the tissue has low UMIs. This suggests that the sample was RNA-poor which could be due to the tissue biology or sample processing methods. 
  • Based on the “Clustering” section of the Web Summary File, the sample was not very complex as there were only a few clusters generated (Figure 3C). The associated t-SNE Projection illustrates that the clustering has very poor association and that spatiality may be compromised (Figure 3D). Poor clustering can be due to tissue biology (e.g., homogenous tissue sample) or poor assay sensitivity leading to the inability to discriminate gene expression differences across the tissue.  

Figure 3. Key Web Summary File plots generated from human bone tissue. (A) Tissue image with (B) UMIs and clustering (C) overlaid or (D) projected. 

  • From the Top Features by Cluster table, there were no differentially expressed genes. Additionally, as the library was at ~98% Sequencing Saturation with only 20,000 Reads per Spot, the library likely consisted of few unique, usable molecules. Moreover, we observed that the Estimated UMIs from Genomic DNA was ~78%, informing us that a substantial portion of reads may be originating from off-target Hybridization to gDNA.

Altogether, across the human bone samples tested we were unable to generate Visium CytAssist Spatial Gene Expression libraries with good quality spatiality, complexity, and sensitivity. Very little usable reads were generated and a high proportion of reads may be attributed to UMIs from gDNA. However, it is important to note that sections from human bone (unlike sections from mouse bone) may not have the same  cellular diversity relative to sample area and this can make differential gene expression analysis challenging. 

In this article, we reviewed our findings and challenges when working with bone samples. Based on limited internal data from healthy mouse femur and tibia FFPE tissue, we found that bone can generate accurate spatial gene expression data. This is shown in Figure 2 where canonical markers of bone (2A), bone marrow (2B), cartilage (2C), and muscle (2D) can be identified in expected regions of the tissue. Thus, it may be possible to generate good-quality Visium data from mouse bone samples by following the 10x Genomics protocols and utilizing Nexterion Hydrogel tissue slides. But additional optimization on targeted regions of mouse bone samples may be required, depending on experimental goals. Contrarily, we determined that human bone samples are more challenging to generate quality gene expression data as even with modifications, spatiality and sensitivity was poor. As data quality may be variable with bone samples, we recommend performing an initial pilot experiment before committing to larger experiments.        

References:

  1. Sighn, V. M., et al., “Analysis of the effect of various decalcification agents on the quantity and quality of nucleic acid (DNA and RNA) recovered from bone biopsies.” Ann Diagn Pathol. 2013; 17(4):322-6. DOI: 10.1016/j.anndiagpath.2013.02.001
  2. Zheng, G., et al., “Clinical Mutational Profiling of Bone Metastases of Lung and Colon Carcinoma and Malignant Melanoma Using Next-Generation Sequencing.” Cancer Cytopathol. 2016;124:744-53. DOI: 10.1002/cncy.21743 

Products: Visium CytAssist for FFPE