How XA v2 pipeline changes impact data comparability

Question: Xenium Analysis v2 software introduces a number of changes from XA-v1. How do the changes impact transcript detection sensitivity and data comparability?

 

Changes in XA-v2 include reductions in channel acquisition exposure times, an updated nuclei segmentation model and reduced maximum isotropic expansion distance from XA-v1. The distance in XA-v1 is 15 µm and in XA-v2 is 5 µm.

 

Answer: All sample preparation and software preprocessing being equal, the exposure time change has minimal impact on transcript detection in two tissue types tested. Furthermore, data remain comparable between v1 and v2 for the two tissue types tested. The two tested tissue types are as follows.

• Fresh-frozen mouse liver with the mouse 379-gene tissue atlas panel 
• FFPE human breast tissue with the 377-gene human multi-tissue and cancer panel 

The fresh-frozen mouse liver tissue and panel combination was most prone to saturation in RNA images, whereas the FFPE human breast tissue and panel combination showed the least saturation across tissue-panel types in internal development. Saturation refers to the per-cycle-per-channel puncta signal intensity.

 

Transcript detection

The XA-2.0 pipeline reduces image acquisition exposure time for two color channels during RCP detection to reduce saturation effects. To assess the impact of the exposure time change, we compared high quality (QV score) transcript detection for the XA-v1.9 pipeline versus XA-2.0 pipeline in the two tissue types.

 

For the comparison, the tissue section uses each exposure setting in interleaved cycles so the v1 versus v2 comparison is from the identical tissue section. The odd numbered cycles use the v2 exposure settings and the even numbered cycles use the v1 exposure settings. DAPI is also re-imaged. Per gene and negative control probe, for QV score >= 20, we calculate gain for each transcript by taking the ratio of v2 to v1 counts and subtracting one. The median gain is the median of these values for each set (genes or negative control probes) after filtering for any transcript with 10 or fewer counts for v1 or v2. This filters 18 transcripts for FFPE human breast (3 genes and 15 NCPs) and filters no transcripts for fresh-frozen mouse liver.

 

Table 1. Transcript detection from XA-v1.9 to XA-v2.0 for QV20+ transcripts. Fraction comparisons use v2 pipeline processed data with the v1 versus v2 exposure times. Although the RNA processing algorithm did not change from v1.9 to v2.0, underlying software package versions did change and may explain minor differences in transcript calling.

Figure1. Transcript counts pseudo-bulk comparison for v1v2 versus v2v2 pipeline runs.

 

Figure 2. Comparison of v1v2 transcript counts against fraction transcript gain from v1v2 to v2v2.

 

UMAP projection and clustering

The exposure time change introduces some minor differences in transcript counts. The differences are pronounced for transcripts with lower expression, e.g. in the same counts range as the negative control probes. To assess for batch effects, we projected PCA reduced FFPE human breast data into the same UMAP space and clustered using KNN nearest neighbors. All three data sets for FFPE human breast (v1v1, v1v2 and v2v2) enable coincident UMAP projection and clustering after filtering cells with no transcripts.

 

Figure 3

   

 

The FFPE human breast v1v2 UMAP projection is more similar to the v2v2 data than to the v1v1 data. The fresh-frozen mouse liver data showed similar concordance (not shown). These results underscore how the exposure time change has less of an effect than the change in default maximum nuclei expansion distance. Furthermore, we can expect greater differences stemming from biological differences between tissue slices from the same block, from different tissue blocks and from different sample preparation batches. 

Table 2. Cell calling and transcript detection from XA-v1.9 to XA-v2.0 pipeline processing for QV20+ transcripts. The XA-v2 pipeline improves the nuclei segmentation model, so the number of cells detected between the v1.9 and v2.0 preprocessing runs differ. In addition, XA-v2 software changes the default maximum nuclei-based free expansion distance to 5 µm. XA-v1 uses 15 µm. This changes the transcripts cells capture from nuclei-based expansion.

Apply XA segmentation to data run on an older pipeline 

It is possible to recapitulate segmentation and nuclei expansion for a pipeline version using Xenium Ranger. Download the latest version from here. For example, to apply the XA v2 pipeline model and expansion parameters to XA v1 data, run Xenium Ranger v2.0 resegment module on the v1 data. The default v2 parameter --expansion-distance=5 will apply.

xeniumranger-xenium2.0/xeniumranger \
resegment \
--xenium-bundle v1bundle \
--id xr2-results \
--expansion-distance 5 \ #default
--localcores 32 \
--localmem 128 

 

Xenium Ranger is not forward-compatible, e.g. running xeniumranger-v1 on XA-v2 data is unsupported and will error.