Evidence of off-target probe binding affecting 10x Genomics Xenium gene panels compromise accuracy of spatial transcriptomic profiling
- Center for Computational Biology, Whiting School of Engineering, Johns Hopkins University, United States
- Department of Biomedical Engineering, Johns Hopkins University, United States
- Department of Computer Science, Johns Hopkins University, United States
- Department of Biostatistics, Johns Hopkins University, United States
Figures
Figure 1
Schematic of potential off-target binding in 10x Genomics Xenium.
In this illustration, the arms of the padlock probes were designed to bind an RNA sequence intended to correspond to a target gene (green). However, these probes exhibit off-target binding and bind to an RNA sequence in a different off-target gene (red). The probe is circularized and subsequently amplified via rolling circle amplification (RCA). Hybridization of fluorescent probes to the RCA product enables the generation of a fluorescent signal that is used to quantify RNA expression within cells.
Figure 2
Comparison of spatial gene expression patterns between Xenium and Visium.
(A) Spatial gene expression of MS4A1 overlaid on the corresponding histological images for Xenium and Visium, accompanied by a density plot comparing Xenium vs. Visium MS4A1 expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (B) Gene expression patterns for APOBEC3B: Xenium expression, Visium expression, the aggregated Visium expression combining APOBEC3B and its predicted off-target gene’s expression APOBEC3D and APOBEC3F, and Visium expression of APOBEC3B’s predicted off-targets APOBEC3D and APOBEC3F. Two density plots are shown: one comparing Xenium vs. Visium for APOBEC3B alone, and one comparing Xenium vs. the aggregated Visium expression of APOBEC3B with all off-targets. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (C) Scatterplot of log-transformed total expression counts (with a pseudocount) for 307 genes comparing Visium and Xenium data. The dotted line indicates the identity line (X = Y), and points (genes) are colored by probe information.
Figure 3
Comparison of single-cell gene expression patterns between Xenium and single-cell RNA sequencing (scRNA-seq).
(A) Harmonized Uniform Manifold Approximation and Projection (UMAP) visualization of MS4A1 expression for Xenium and scRNA-seq data, accompanied by a scatterplot comparing Xenium vs. scRNA-seq MS4A1 cluster expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (B) Comparison of APOBEC3B expression patterns on harmonized UMAP: Xenium expression, scRNA-seq expression, an aggregated scRNA-seq profile combining APOBEC3B and its predicted off-target genes’ expression APOBEC3D and APOBEC3F, and scRNA-seq expression of APOBEC3B’s predicted off-targets APOBEC3D and APOBEC3F. Two scatterplots are shown: one comparing Xenium vs. scRNA-seq for APOBEC3B cluster expression alone, and one comparing Xenium vs. the aggregated scRNA-seq cluster expression of APOBEC3B and its predicted off-targets. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (C) Scatterplot of log-transformed total expression counts (with a pseudocount) for 313 genes between Visium and scRNA-seq data. The dotted line indicates the identity line (X = Y), and points (genes) are colored by probe information.
Appendix 1—figure 1
Screenshot from the Integrated Genome Viewer (IGV) showing the 40 bp probe target sequence (ID: ENSG00000196154|S100A4|ab4e3dc) that matches both S100A5 and S100A4.
Shown are six isoforms from the CHESS v3.1 annotation, four from GENCODE basic v47, and three from RefSeq v110 for S100A4, as well as two RefSeq isoforms for the neighboring S100A5. The probe target sequence aligns to the overlapping region between S100A5 and S100A4 gene loci. Matching probe shown in a zoomed-in view below. The forward- and reverse-strand sequences of the probe are shown, and the highlighted area indicates approximately where the probe falls within the gene.
Appendix 1—figure 2
UpSet plot illustrating the overlap of protein-coding genes across three genome annotations: GENCODE basic, RefSeq, and CHESS.
Appendix 1—figure 3
Preprocessing the Visium and Xenium datasets.
(A) Overlap regions between Visium (orange outline) and Xenium (blue outline) data, shown on the Xenium histological image and the Visium histological image, respectively. (B) Log transformed aggregated total gene counts for spots (~55 μm × 55 μm) in both Xenium and Visium datasets, overlaid on their corresponding histological image.
Appendix 1—figure 4
Screenshot from the Integrated Genome Viewer (IGV) showing the 40 bp probe sequences that matches both APOBEC3B as well as APOBEC3D and APOBEC3F.
(A) Four 40 bp probes targeting APOBEC3B (ENSG00000179750|APOBEC3B|17d76bb, ENSG00000179750|APOBEC3B|e03f8ab, ENSG00000179750|APOBEC3B|5991db9, and ENSG00000179750|APOBEC3B|59c9349). (B) All four probe target sequences align to their intended target gene APOBEC3B, while two of the four probe target sequences align to each off-target gene: (C) APOBEC3D and (D) APOBEC3F. The forward- and reverse-strand sequences of the probe target sequences are shown, and the highlighted areas indicate approximately where the probe target sequence falls within the gene. Panels (B–D) share a common legend.
Appendix 1—figure 5
Uniform Manifold Approximation and Projection (UMAP) visualization of integrated single-cell RNA sequencing (scRNA-seq) and Xenium datasets.
(A) Before harmony batch correction and (B) after harmony batch correction. (C) Leiden clustering results on the harmonized UMAP.
Appendix 1—figure 6
Illustrative schematics of potential probe binding issues.
(A) Schematic illustrating that hybridization may still occur even when there is a sequence mismatch at the non-ligated ends of the probe target sequence. (B) Schematic depicting how probes could bind to each other instead of to their intended target.
Appendix 1—figure 7
Effect of Predicted Off-Target Probe Binding on ACTG2 and TUBB2B Expression Patterns using Visium data.
(A) Gene expression patterns for ACTG2: Xenium expression, Visium expression, the aggregated Visium expression combining ACTG2 and its predicted off-target gene’s expression ACTA1, ACTB, and POTEM, and Visium expression of ACTG2’s predicted off-targets ACTA1, ACTB, and POTEM. Two density plots are shown: one comparing Xenium vs. Visium for ACTG2 alone, and one comparing Xenium vs. the aggregated Visium expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (B) Gene expression patterns for TUBB2B: Xenium expression, Visium expression, the aggregated Visium expression combining TUBB2B and its predicted off-target gene’s expression TUBB2B and TUBB2A, and Visium expression of TUBB2B’s predicted off-target TUBB2A. Two density plots are shown: one comparing Xenium vs. Visium for TUBB2B alone, and one comparing Xenium vs. the aggregated Visium expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit.
Appendix 1—figure 8
Effect of Predicted Off-Target Probe Binding on ACTG2 and TUBB2B Expression Patterns using scRNA-seq data.
(A) Comparison of ACTG2 expression patterns on harmonized Uniform Manifold Approximation and Projection (UMAP): Xenium expression, single-cell RNA sequencing (scRNA-seq) expression, an aggregated scRNA-seq profile combining ACTG2 and its predicted off-target gene’s expression ACTB, POTEM, POTEE, POTEF, POTEI, POTEJ, and ACTA1, and scRNA-seq expression of ACTG2’s potential off-targets. Two scatterplots are shown: one comparing Xenium vs. scRNA-seq for ACTG2 cluster expression alone, and one comparing Xenium vs. the aggregated scRNA-seq cluster expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (B) Comparison of TUBB2B expression patterns on harmonized UMAP: Xenium expression, scRNA-seq expression, an aggregated scRNA-seq profile combining TUBB2B and its predicted off-target gene’s expression TUBB2A, and scRNA-seq expression of TUBB2B’s potential off-target TUBB2A. Two scatterplots are shown: one comparing Xenium vs. scRNA-seq for TUBB2B cluster expression alone, and one comparing Xenium vs. the aggregated scRNA-seq cluster expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit.
Appendix 1—figure 9
Heatmap visualization of target genes and their predicted off-target genes of the HuBMAP placenta custom probe panel using a corresponding placenta bulk RNA-seq dataset.
Gene expression values are counts per million (CPM) normalized, with a pseudocount added prior to log transformation. The first column shows the expression of each target gene in the placenta bulk RNA-seq dataset, while the remaining columns display the expression of the corresponding predicted off-target genes.
Appendix 1—figure 10
Heatmap visualizations of target genes and their predicted off-target genes of the HuBMAP multi custom probe panel using a corresponding (A) kidney, (B) lung, and (C) heart single-cell RNA sequencing (scRNA-seq) datasets.
Gene expression values are counts per million (CPM) normalized, with a pseudocount added prior to log transformation. The first column shows the expression of each target gene in their respective scRNA-seq dataset, while the remaining columns display the expression of the corresponding predicted off-target genes.
Appendix 1—figure 11
Heatmap visualizations of target genes and their predicted off-target genes for the Janesick et al. probes using (A) pseudo-bulked single-cell RNA sequencing (scRNA-seq) and (B) Visium data.
Gene expression values are counts per million (CPM) normalized, with a pseudocount added prior to log transformation. The first column shows the expression of each target gene in their respective scRNA-seq dataset, while the remaining columns display the expression of the corresponding predicted off-target genes.
Appendix 1—figure 12
Effect of Predicted Off-Target Probe Binding on ADH1B Expression Patterns using Visium and scRNA-seq data.
(A) Gene expression patterns for ADH1B: Xenium expression, Visium expression, the aggregated Visium expression combining ADH1B and its predicted off-target gene’s expression ADH1A and ADH1C, and Visium expression of ADH1B’s predicted off-targets ADH1A and ADH1C. Two density plots are shown: one comparing Xenium vs. Visium for ADH1B alone, and one comparing Xenium vs. the aggregated Visium expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit. (B) Comparison of ADH1B expression patterns on harmonized Uniform Manifold Approximation and Projection (UMAP): Xenium expression, single-cell RNA sequencing (scRNA-seq) expression, an aggregated scRNA-seq profile combining ADH1B and its predicted off-target gene’s expression ADH1A and ADH1C, and scRNA-seq expression of ADH1B’s potential off-targets ADH1A and ADH1C. Two scatterplots are shown: one comparing Xenium vs. scRNA-seq for ADH1B cluster expression alone, and one comparing Xenium vs. the aggregated scRNA-seq cluster expression. The dotted line indicates the identity line (X = Y), and the solid line represents the line of best fit.
Appendix 1—figure 13
Spatial gene expression of HDC in Xenium, Visium, and scRNA-seq data.
(A) Spatial gene expression of HDC overlaid on the corresponding histological images for Xenium and Visium. (B) Harmonized Uniform Manifold Approximation and Projection (UMAP) visualization of HDC expression for Xenium and single-cell RNA sequencing (scRNA-seq) data.
Appendix 1—figure 14
Screenshots from the Integrated Genome Viewer (IGV) illustrating annotation-dependent differences in probe alignment.
(A) A 40-bp probe (ID: ENSG00000125878|TCF15|5d3cbc2) aligns to an exonic region in GENCODE v28 but to an upstream region in GENCODE v47. (B) A 40-bp probe (ID: ENSG00000169083|AR|a0c6719) aligns to an exonic region in GENCODE v28 but to an intronic region in GENCODE v47. Matching probe shown in a zoomed-in view below. The forward- and reverse-strand sequences of the probe are shown, and the highlighted areas indicate approximately where the probe falls within the gene. Panels (A, B) share a common legend.
Tables
Table 1
Off-target Probe Tracker (OPT) output of genes with predicted off-target binding based on perfect sequence homology using GENCODE v47.
This table shows the 37 genes whose probes in the 10x Genomics Xenium v1 Human Breast Gene Expression Panel exhibit predicted off-target probe binding, where each off-target alignment involves a perfect 40 bp match to the probe target sequence. Although OPT predicted off-target binding of CCPG1 probe target sequences to the DNAAF1-CCPG1 gene, we manually excluded it from our list because DNAAF1-CCPG1 is a read-through gene containing portions of both DNAAF1 and CCPG1. The final column shows the gene types, in order, of each of the off-target genes shown in column 3. Abbreviations: PC = protein-coding; PG = pseudogene; NMD = nonsense-mediated decay; lncRNA = long non-coding RNA.
| Target gene | Number of probes | Predicted binding genes | Number of probes aligned | Gene types – GENCODE (v47) |
|---|---|---|---|---|
| ADH1B | 8 | ADH1B, ADH1A, ADH1C | 8, 2, 1 | PC, PC, PC |
| AKR1C1 | 9 | AKR1C1, AKR1C2, AKR1C3, AKR1C4, AKR1C5P | 9, 1, 1, 1, 1 | PC, PC, PC, PC, PG |
| APOBEC3A | 8 | APOBEC3A, APOBEC3B | 8, 2 | PC, PC |
| APOBEC3B | 8 | APOBEC3B, APOBEC3D, APOBEC3F, ENSG00000284554 | 8, 2, 2, 2 | PC, PC, PC, PC |
| AQP1 | 10 | AQP1, ENSG00000250424 | 10, 4 | PC, PC |
| C15orf48 | 6 | C15orf48, MIR147B | 6, 1 | PC, miRNA |
| C1QA | 4 | C1QA, ENSG00000289692 | 4, 2 | PC, PC |
| CD68 | 7 | CD68, ENSG00000264772 | 7, 6 | PC, lncRNA |
| CD79B | 5 | CD79B, ENSG00000285947 | 5, 3 | PC, PC |
| CD8B | 16 | CD8B, CD8B2 | 16, 2 | PC, PC |
| CEACAM6 | 8 | CEACAM6, ENSG00000267881 | 8, 1 | PC, PC |
| CLECL1; CLECL1P | 3 | CLECL1P, ENSG00000293488 | 3, 3 | PG, lncRNA |
| DPT | 8 | DPT, LINC00970 | 8, 8 | PC, lncRNA |
| EPCAM | 8 | EPCAM, ENSG00000225356 | 8, 1 | PC, PG |
| HMGA1 | 7 | HMGA1, HMGA1P1, HMGA1P2, HMGA1P3 | 7, 1, 1, 1 | PC, PG, PG, PG |
| IL2RG | 9 | IL2RG, ENSG00000285171 | 9, 8 | PC, NMD |
| KRT14 | 6 | KRT14, KRT16P6, ENSG00000290977 | 6, 1, 1 | PC, PG, lncRNA |
| KRT8 | 16 | KRT8, KRT8P3, KRT8P2, KRT8P33, KRT8P45, CDK5R2-AS1, ENSG00000304440, KRT8P11, KRT8P17, KRT8P22, KRT8P30, KRT8P32, KRT8P36, KRT8P37, KRT8P42 | 16, 3, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 | PC, PG, PG, PG, PG, lncRNA, lncRNA, PG, PG, PG, PG, PG, PG, PG, PG |
| LDHB | 8 | LDHB, ENSG000002854 | 8, 5 | PC, NMD |
| LILRA4 | 8 | LILRA4, ENSG00000275210 | 8, 1 | PC, lncRNA |
| MYLK | 11 | MYLK, MYLKP1 | 11, 1 | PC, PG |
| MYO5B | 8 | MYO5B, MYO5BP1, MYO5BP2, | ||
| ENSG00000266997 | 8, 1, 1, 4 | PC, PG, PG, NMD | ||
| PCLAF | 8 | PCLAF, ENSG00000259316 | 8, 1 | PC, NMD |
| POLR2J3 | 10 | POLR2J3, POLR2J4, POLR2J, ENSG00000270249, POLR2J2, POLR2J2-UPK3BL1, ENSG00000291154 | 10, 4, 3, 2, 2, 2, 1 | PC, lncRNA, PG, PC, PC, PC, NMD, lncRNA |
| PTGDS | 5 | PTGDS, ENSG00000284341 | 5, 3 | PC, NMD |
| SCD | 8 | SCD, SCDP1 | 8, 2 | PC, PG |
| SERHL2 | 8 | SERHL2, SERHL | 8, 7 | PC, PG |
| SERPINA3 | 8 | SERPINA3, ENSG00000273259 | 8, 8 | PC, NMD |
| SLAMF1 | 10 | SLAMF1, ENSG00000228863 | 10, 1 | PC, lncRNA |
| SMS | 8 | SMS, ENSG00000213080, ENSG00000232389, ENSG00000249779 | 8, 3, 1, 1 | PC, PG, PG, PG |
| THAP2 | 13 | THAP2, ENSG00000258064 | 13, 2 | PC, NMD |
| TPD52 | 8 | TPD52, ENSG00000276418 | 8, 5 | PC, NMD |
| TPSAB1 | 2 | TPSAB1, TPSB2, TPSD1 | 2, 2, 1 | PC, PC, PC |
| TRAF4 | 9 | TRAF4, ENSG00000225869 | 9, 1 | PC, PG |
| TUBB2B | 8 | TUBB2B, TUBB2BP1 | 8, 1 | PC, PG |
| VOPP1 | 11 | VOPP1, ENSG00000223612 | 11, 1 | PC, PG |
| VWF | 8 | VWF, VWP1 | 8, 1 | PC, PG |
Additional files
-
Supplementary file 1
Fasta file of probe target sequences from 10x Genomics corresponding to the 10x Genomics Xenium v1 Human Breast Gene Expression Panel used in Janesick et al.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp1-v1.zip
-
Supplementary file 2
Off-target Probe Tracker (OPT) output of genes with predicted off-target binding based on perfect sequence homology in RefSeq.
This table shows the 14 genes whose probes in the 10x Genomics Xenium v1 Human Breast Gene Expression Panel exhibit predicted off-target probe binding, where each off-target alignment involves a perfect 40 bp match to the probe target sequence. The final column shows the gene types, in order, of each of the off-target genes shown in column 3. Off-target alignments between CCPG1 probes and DNAAF1-CCPG1 were excluded. Abbreviations: PC = protein-coding; PG = pseudogene; precursor_RNA = precursor RNA; misc_RNA = miscellaneous RNA; ncRNA = non-coding RNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp2-v1.xlsx
-
Supplementary file 3
Off-target Probe Tracker (OPT) output of genes with predicted off-target binding based on perfect sequence homology in CHESS.
This table shows the 23 genes whose probes in the 10x Genomics Xenium v1 Human Breast Gene Expression Panel exhibit predicted off-target probe binding, where each off-target alignment involves a perfect 40 bp match to the probe target sequence. The final column shows the gene types, in order, of each of the off-target genes shown in column 3. Off-target alignments between CCPG1 probes and DNAAF1-CCPG1 were excluded. Abbreviations: PC = protein-coding; PG = pseudogene; miRNA = microRNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp3-v1.xlsx
-
Supplementary file 4
The number of off-target probes and affected genes (from the set of 313 genes in the Xenium panel) found when looking for perfect matches between probe target sequences and transcripts in four different reference annotations: GENCODE basic, GENCODE comprehensive, RefSeq, and CHESS.
Off-target alignments between CCPG1 probes and DNAAF1-CCPG1 were excluded.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp4-v1.xlsx
-
Supplementary file 5
Union set of protein-coding genes that Off-target Probe Tracker (OPT) predicts to be affected by off-target binding, across three different reference annotations: GENCODE basic, RefSeq, and CHESS.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp5-v1.xlsx
-
Supplementary file 6
Eighteen additional genes that were identified to exhibit potential off-target probe binding when allowing for a 10-bp error margin on either side of the binding site when using GENCODE v47.
Abbreviations: PC = protein-coding; PG = pseudogene; lncRNA = long non-coding RNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp6-v1.xlsx
-
Supplementary file 7
Ten genes that were previously predicted to be affected by off-target binding based on perfect matching now show additional predicted off-target interactions when a 10-bp error margin is allowed on either side of the binding site when using GENCODE v47.
This effect is observed either through the accumulation of new probes with predicted off-target binding or via the identification of additional predicted off-target genes per probe. Abbreviations: PC = protein-coding; PG = pseudogene; lncRNA = long non-coding RNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp7-v1.xlsx
-
Supplementary file 8
Off-target Probe Tracker (OPT) output for 49 genes with predicted off-target binding in the HuBMAP placenta custom probe panel, generated using GENCODE v47 and allowing a 10-bp mismatch on either end of each probe.
The final column lists the gene types, in order, corresponding to the off-target genes shown in column 3. Abbreviations: PC = protein-coding; PG = pseudogene; precursor_RNA = precursor RNA; misc_RNA = miscellaneous RNA; ncRNA = non-coding RNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp8-v1.xlsx
-
Supplementary file 9
Off-target Probe Tracker (OPT) output for 24 genes with predicted off-target binding in the HuBMAP multi custom probe panel, generated using GENCODE v47 and allowing a 10-bp mismatch on either end of each probe.
The final column lists the gene types, in order, corresponding to the off-target genes shown in column 3. Abbreviations: PC = protein-coding; PG = pseudogene; precursor_RNA = precursor RNA; misc_RNA = miscellaneous RNA; ncRNA = non-coding RNA.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp9-v1.xlsx
-
Supplementary file 10
Off-target Probe Tracker (OPT) results for all publicly available 10x Genomics probe sets.
Results include all possible RNA species, and using a pad length (-pl) of 10.
- https://cdn.elifesciences.org/articles/107070/elife-107070-supp10-v1.zip
-
MDAR checklist
- https://cdn.elifesciences.org/articles/107070/elife-107070-mdarchecklist1-v1.docx
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Evidence of off-target probe binding affecting 10x Genomics Xenium gene panels compromise accuracy of spatial transcriptomic profiling
eLife 14:RP107070.
https://doi.org/10.7554/eLife.107070.3
