[ 공간전사체 논문 ] FFPE 이용 가능한 공간전사체 툴 비교/밴치마킹 (10x Xenium vs Vizgen MERSCOPE vs Nanostring(bruke) CosMX) "Systematic benchmarking of imaging spatial transcriptomics platforms in FFPE tissues" 2023.12

 

⇒ OVERALL

Cell Segmentation 은 지금 2025/02에는 또 달라졌을 수도 있지만 해당 페이퍼에서는 CosMx가 Segmentation은 잘하지만, Cell type recovery based on gene markers는 Xenium이 확실히 뛰어난다. 만약 샘플이 오래되어서 RNA degradation 이 심하면 Xenium와 같이 fewer landing sites, heavily amplified 되는 프로레스가 가장 적합하다. 샘플 퀄리티가 좋으면 어떤 플랫폼이든 비슷한 결과를 보인다.참고로 최근에 'proseg'라고 segmentation하는 툴 나온게 있는데 확인해볼만 하다.

https://github.com/dcjones/proseg 

Spatial Transcriptomics (ST) tools

• sequencing-based (sST): place the tissue on barcoded substrate
• image-based (iST): barcoded probe hybridization
  • 🙂 Higher spatial resolution and sensitivity of FISH

Sample PreP:

• the need of amplification depends on sample processing, cleared, gel-embedded or photo-bleached
• tradeoffs between imaging time, molecular plex, image are covered
• clearing of the sample → ⬆️ signal quality ❌ follow-up H&E staining or immunostaining = make cell segmentation more challenging
  
  **FFPE (formalin-fixed paraffin-embedded): standard format for clinical sample preservation for pathology due to its ability to maintain tissue morphology and sample stability at room temperature for decades.

 

Xenium vs MERSCOPE vs CosMx

• different protocols, probe designs, signal amplification strategies and computational processing methods

	Xenium	MERSCOPE	CosMx
sample prep	2-3 days	5-7 days, cutting sample onto MERSCOPE coverslip is a bit difficult	2 days
batch processing	🤔	🙂	🙂
Transcript Amplification	a small number of padlock probes with rolling circle amplification	direct probe hybridization but amplifies by tiling the transcript with many probes	a low number of probes amplified with branch chain hybridization
Transcripts counts	High	.	High
Transcript counts per gene	High	.	.
Cell Segmentation	DAPI → bigger areas	DAPI + membrane markers → tighter which resulting removing more transcripts but more confidently assigned to cells
Cell Type recovery	🙂	~	atypical gene markers and low expression of canonical markers,
Study	better suited for branches of biology not well sampled by the 1,000 plex CosMx panel		the higher false positive rates and lower sensitivities of CosMx relative to Xenium could be tolerated for a broader coverage of the biology.

 

RESULTS

DATA: 23 FFPE tissue types (7 tumor, 16 normal), > 3.3 M cells, > 190 million transcripts

EXPERIMENTs:

1. sensitivity and specificity on shared transcripts
2. concordance of the iST data
3. cell-level comparisons: evaluating segmentation
4. ability to identify cell-type clusters

RESULTs:

• High transcript counts obtained by Xenium and CosMx
• Xenium shows higher transcript counts per gene without sacrificing specificity
• iST platforms are all concordant with orthogonal RNA-seq data sets
• Out of the box segmentation - Xenium은 DAPI(nuclei staining)로만 segmentation을 해서 빈공간들을 세포로 잡아버린다. 
• Clustering analysis reveals differences in cell type recovery across platforms

⇒ OVERALL

Cell Segmentation 은 지금 2025/02에는 또 달라졌을 수도 있지만 해당 페이퍼에서는 CosMx가 Segmentation은 잘하지만, Cell type recovery based on gene markers는 Xenium이 확실히 뛰어난다. 만약 샘플이 오래되어서 RNA degradation 이 심하면 Xenium와 같이 fewer landing sites, heavily amplified 되는 프로레스가 가장 적합하다. 샘플 퀄리티가 좋으면 어떤 플랫폼이든 비슷한 결과를 보인다.참고로 최근에 'proseg'라고 segmentation하는 툴 나온게 있는데 확인해볼만 하다.

https://github.com/dcjones/proseg