Help interpreting GSEA results

[Thorben Sauer]

Hey,

I am currently doing a proteomics based characterization of SW480 and SW620 colorectal cancer cell lines as part of a larger project. For this, I applied GSEA using the R package clusterprofiler on the MSigDB::C2 gene set collection. Not very surprisingly, the gene sets "PROVENZANI_METASTASIS_UP" and "PROVENZANI_METASTASIS_DOWN" popped up, which describe "Genes up/down-regulated in polysomal and total RNA samples from SW480 cells (primary colorectal carcinoma, CRC) compared to the SW620 cells (lymph node metastasis from the same individual)." So far so good.

However, I get a very strong enrichment for the DOWN Geneset for SW480 and the UP Geneset for SW620 and vice versa, which in my opinion, does not make sense. To further investigate and to check, if I made a coding mistake or similar, I checked a couple logFC for corresponding genes from both gene sets. Most genes I checked of the DOWN geneset were actually upregulated in SW480 in our dataset. In line, most genes of the UP regulated gene set were down regulated in SW620 in our set. For these genes, I further checked the expression pattern for cell lines reported in the protein atlas. All the genes of the DOWN geneset I checked showed higher expression in SW480 compared to SW620, and vice versa for the genes of the UP geneset I checked . A very prominent example is the gene ANXA6, which is listed in the DOWN geneset but shows largely higher expression in SW480 (https://www.proteinatlas.org/ENSG00000197043-ANXA6/cell+line).

To go one step further, I checked the Geo accession of the corresponding dataset (https://www.ncbi.nlm.nih.gov/geo/geo2r/?acc=GSE2509). Again, the differential expression analysis tool provided by GEO indicates higher expression levels of ANXA6 in SW480, although it is listed in the geneset PROVENZANI_METASTASIS_DOWN, which contains proteins that are downregulated in SW480 compared to SW620.

Am I totally misunderstanding the description of this geneset or is it actually misleading or even switched up?

Kind regards,

Thorben

[Castanza, Anthony]

Hi Thorben,

Unfortunately it appears as if the full supplementary data for this paper is no longer available, however, a selection of qPCR validated genes was available in the manuscript text, I was able to confirm that their top hits KARS (now KARS1), AARS (now AARS1), and TARS (now TARS1) are present as expected in the PROVENZANI_METASTASIS_UP signature – so, our signature annotation does appear to match was was presented in the original publication – at least, as much of it as is currently accessible.

 

Original Publication (note KARS)

However, looking deeper at the GEO data, I do, in fact, see what you see here:

GEO Data (KARS, same probe as MSigDB UP Signature):

I would have been inclined to suggest that this was an error in the submission to GEO where the annotation was inadvertently switched, but in light of your additional analysis I think the issue here might be slightly more insidious – that the phenotypes were inadvertently switched in the original published analysis.

Thank you for bringing this to our attention. I’m CC’ing the author of the original publication here (Alessandro Provenzani), perhaps they have some additional insight on what might have happened here and what, if any, corrections might be possible.

-Anthony

 

Anthony S. Castanza, PhD

Department of Medicine

University of California, San Diego

--
You received this message because you are subscribed to the Google Groups "gsea-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gsea-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/721575c0-ad13-4929-94ec-6cbf7fec91e2n%40googlegroups.com.

 

[Castanza, Anthony]

Hi Thorben,

Unfortunately it appears as if the full supplementary data for this paper is no longer available, however, a selection of qPCR validated genes was available in the manuscript text, I was able to confirm that their top hits KARS (now KARS1), AARS (now AARS1), and TARS (now TARS1) are present as expected in the PROVENZANI_METASTASIS_UP signature – so, our signature annotation does appear to match was was presented in the original publication – at least, as much of it as is currently accessible.

 

Original Publication (note KARS)

However, looking deeper at the GEO data, I do, in fact, see what you see here:

GEO Data (KARS, same probe as MSigDB UP Signature):

I would have been inclined to suggest that this was an error in the submission to GEO where the annotation was inadvertently switched, but in light of your additional analysis I think the issue here might be slightly more insidious – that the phenotypes were inadvertently switched in the original published analysis.

Thank you for bringing this to our attention. I’m CC’ing the author of the original publication here (Alessandro Provenzani), perhaps they have some additional insight on what might have happened here and what, if any, corrections might be possible.

-Anthony

 

Anthony S. Castanza, PhD

Department of Medicine

University of California, San Diego

(Resent due to address failure)

[Thorben Sauer]

Hey Anthony,

thank you for looking into this. I noted the inavailability of the supplementary data as well, which is why I tried to look into the GEO data directly. Hopefully, there is a solution to this.

Best

Thorben

[Thorben Sauer]

Dear Anthony,

did you get any response from the Mr. Provenzani regarding this issue? How would MSigDB handle the corresponding dataset, assuming you get no response from the authors?

Best

Thorben

[Castanza, Anthony]

Hi Thorben

No, I haven’t heard anything back from the author. If we’re unable to get a clarification, I’m currently inclined to withdraw the set however we haven’t made a final determination yet. It may be possible to rework the set from just the GEO data in a satisfactory way.

 

-Anthony

 

Anthony S. Castanza, PhD

Department of Medicine

University of California, San Diego

 

To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/311ee0c2-4729-4195-99c1-ea1a3d0b2acan%40googlegroups.com.