Can Chromium Single Cell Gene Expression Flex interrogate patient-derived xenograft (PDX) samples?

Question:

Can Chromium Single Cell Gene Expression Flex interrogate patient-derived xenograft (PDX) samples?

Answer:

Patient-derived xenograft (PDX) samples typically contain human tumor tissue grown in an immunodeficient or humanized recipient specimen (e.g. mouse). Xenograft tissue more accurately reflects the biological characteristics of the cancer compared to cell-line models. However, xenografted tissue is often vascularized by recipient cells, and this contamination can significantly impact downstream analysis if not removed.

Cell Ranger v9.0 enables multi-species analysis with Flex, allowing the detection and subsequent removal of mouse barcodes from analysis. We offer two methods to remove recipient cells from analysis of PDX samples.

1. Spike in custom probes: Add custom probes that are specific to the recipient species to unambiguously identify recipient cells in Loupe Browser. Once identified, remove these barcodes from subsequent analysis. We provide a recommended set of spike-in probes for mouse, as well as a general strategy for developing probes for other recipient species.
2. Probe set reference analysis: Analyze samples in Cell Ranger using a probe-set reference that lacks probes expected to cross react with the recipient transcriptome. While this method may filter out some true human cells,  it does not require the purchase of additional probes. The identified human barcodes can subsequently be rerun through Cell Ranger using the full human probe-set reference.

The xenograft samples illustrated below are:

1. 786-O human renal cell carcinoma cell-derived tumor tissue grown in SCID-beige mice.
2. H-358 human non small cell lung carcinoma grown in NSG mice.

Method 1 - Spike in custom species-specific probes

Workflow

1. Order probes following guidance from “Custom Probe Design for Visum Spatial Gene Expression and Chromium Single Cell Gene Expression Flex” (CG000621). A list of recommended mouse probe sequences used for human:mouse xenografts is provided here. Note: Probe barcode (BC001) is used in this list. Ensure that the probe barcode on the custom spike-in matches the probe barcode on the human whole transcriptome analysis (WTA) probe used in the hybridization. 
2.  Generate fixed cells from the human-mouse xenograft sample by following the appropriate protocol:
   • GEM-X: “Tissue Fixation & Dissociation for GEM-X Flex Gene Expression” (CG000783)

           OR

• • Next GEM: “Tissue Fixation & Dissociation for Fixed RNA Profiling” (CG000553).

3. Set up probe hybridization with spike-in probes following “Custom Probe Design for Visium Spatial Gene Expression and Chromium Single Cell Gene Expression Flex” (CG000621). 

Analysis

The use of custom probes requires modifications to the Probe Set Reference CSV. A probe set CSV compatible with the mouse-spike in probes used in this demonstration is provided here. Probes in the final panel target Sdha, B2m, Gusb, Malat1, Rpl41, and Eef1a1 across 22 total probe pairs. If you are using different probes or targeting a different recipient species, please refer to the modifications described in the Custom Probe Design for Visium Spatial Gene Expression and Chromium Single Cell Gene Expression Flex (CG000621).

Within the Loupe Browser, expression of the summation of mouse transcripts (GRCm39 Sum) can be used to select mouse barcodes (Figure 1). Cell Ranger can then be re-run using customized secondary analysis to exclude mouse barcodes and generate data representing the human sample only.

(A)                                                                                                (B)

Figure 1:  t-SNE plots from Loupe Browser plotting the log2 sum expression of mouse spike-in probes (GRCm39 Sum) in 786-O xenografts (A). Clustering and expression of mouse probes enabled identification of mouse (purple) and human (grey) cells (B).

Considerations for designing probes targeting non-mouse recipients

When targeting probes for xenograft analysis, it is generally preferred to target moderately to highly expressed genes. For mouse, candidate genes were identified using the Tabula muris dataset. It is recommended to target approximately 10-20 genes, each with ~8 probes per gene. Follow guidelines and considerations in outlined in the Custom Probe Design for Visium Spatial Gene Expression and Chromium Single Cell Gene Expression Flex (CG000621) Technical Note when designing probes for xenograft analysis. Additionally, use Basic Local Alignment Search Tool (BLAST) to check for off target activity of candidate probes.

Initial examination of expression and specificity of spike-in probes can be performed in Loupe Browser, but further investigation at the per-probe level is recommended. The probe level count matrix can be found under per_sample_outs folder in Cell Ranger output (sample_raw_probe_bc_matrix.h5) and can be parsed using common third-party tools including Seurat and Scanpy. Analyze at a pseudobulk level at increasing K-means clustering until one cluster is identified having the majority of counts for the spike-in probes. Label this cluster as the recipient cluster and group the remaining clusters into a human cluster. Generate a table of pseudobulk mean counts per cell for each species, and filtered to the spike-in probes. Manually inspect to identify probes with zero (or near-zero) counts in the human cluster and several in the recipient cluster. Computationally exclude unsatisfactory probes from subsequent analysis. An example table from our development of mouse spike-in probes is provided below. 

Table 1: Example human and mouse counts

Method 2 - Use a reference lacking probes expected to cross-react with the recipient’s transcriptome

Workflow

1. Process the PDX sample to generate fixed cells following:
   • Next GEM: “Tissue Fixation & Dissociation for Fixed RNA Profiling” (CG000553).

           OR

1. • GEM-X: “Tissue Fixation & Dissociation for GEM-X Flex Gene Expression” (CG000783).
2. Set up probe hybridization as normal with the Human WTA probe set.
3. Run through Cell Ranger using filtered reference lacking probes expected to cross-react with mouse.
4. Recommended: Compare to Cell Ranger output using standard probeset. Manually confirm that removed (recipient) barcodes are correct. Adjust if necessary.
5. Re-run human-specific barcodes through Cell Ranger using standard reference.

Considerations

An alternative to the probe spike-in approach, that does not require purchasing additional probes, is to use the Human WTA panel only. However, the homology between the human and mouse transcriptomes leads to high cross-reactivity between human probes and mouse genes making it difficult to define mouse barcodes with high confidence. Mapping reads to a filtered reference lacking probes expected to cross react with mouse may be a viable work around.

To prepare the modified target panel, Basic Local Alignment Search Tool (BLAST) was used to identify probes from the Human WTA panel that are expected to cross react with the mouse transcriptome. A human probe is computationally predicted to be cross reactive if the RHS & LHS probes have a Tm (melting temperature) of >60℃ and there is no mismatch at the 25th nucleotide (with the target) when BLASTed against the transcriptome of other species. Approximately 57% of all Human WTA probes were computationally removed due to elevated levels of expected cross reactivity. The filtered probeset is here.  

Probe melting temperatures were calculated by primer3 (https://primer3.org). The specific parameters used in primer3 are shown below. These parameters can be executed by calling

primer3.calc _ tm(probe _ seq,
**PRIMER3 _ CALCTM _ ARGS).

PRIMER3 _ CALCTM _ ARGS = {
    "mv _ conc": 10,
    "dv _ conc": 20,
    "dntp _ conc": 10,
    "dna _ conc": 5000,
    "salt _ corrections _ method":
"schildkraut"
}

H-358 xenografts were processed with Human WTA probes only and run through Cell Ranger with either the standard reference or the filtered reference that is predicted to not map to the mouse transcriptome. To illustrate which barcodes were lost upon mapping to filtered reference, the excluded barcodes were plotted on the original t-SNE from the standard reference. An isolated cluster (Figure 2C, circled) was identified that is likely composed of mouse cells, but additional (presumably human) barcodes in the main cluster were also removed. These putative human barcodes were likely lost from this analysis because removal of probes from the reference reduced the UMI counts for these barcodes to a level below our cell calling threshold.

The manual interrogation of lost barcodes as described above is recommended, as a non-insignificant amount of human cells would be otherwise removed from analysis. For example, ~14% of human alveolar type 1 (AT1) cells would be lost from the non-small cell lung carcinoma xenograft without manual interrogation.

Similar to Method 1, upon identification of human barcodes in Loupe, human barcodes can be re-run in Cell Ranger using the standard human reference.

(A)                                                      (B)                                                                         (C)

Figure 2. t-SNE plots from Loupe Cell Browser for H-358 non-small cell lung carcinoma are shown when mapped to either the standard reference (A) or filtered reference lacking probes expected to cross react with mouse (B). Barcodes excluded from analysis when using the filtered reference were plotted on the t-SNE from the standard reference (C, purple). Putative mouse cells in the H-358 xenograft sample are circled in red.

Disclaimer:

At this time, PDX models are not supported with any of our assays and the performance of these methods is not guaranteed. The use of custom probes in these assays is not officially supported by 10x Genomics. The additional guidance and resources that have been provided in this article are intended to help enable these experiments and improve customer success.