HOOK2 opposite effect than PAFAH1B1, NDE1, NDEL1: exploration for MorphMap paper (ORF)

AnneCarpenter commented 9 months ago

We found a tight cluster of PAFAH1B1, NDE1, NDEL1, and those 3 have the opposite morphological effect to the gene HOOK2. Emailed Sivaram V.S. Mylavarapu Nov 29, 2023: https://mail.google.com/mail/u/0/#sent/KtbxLwgxBKKHZtcxwBdZCLKMjxlVNDmDqV No reply so then I emailed Jan 5: msharma@iisermohali.ac.in because at least they studied HOOK2: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6400558/

AnneCarpenter commented 9 months ago

Info provided here now: https://github.com/broadinstitute/2023_12_JUMP_data_only_vignettes/issues/8

tjetkaARD commented 8 months ago

Claryfing the results:

similarly as for RAB40B and PIK3R3 - it is a result only for ORFs - it cannot be confirmed within CRISPRs morphology

Nonethless, the underlying biology seems to be quite consistent with literature:

Consistent overall negative co-expression between HOOK2 and cluster of: PAFAH1B1, NDE1, NDEL1, with very interesting tissue-level heterogeneity:
- e.g. PAFAH1B1 is amongh top 0.5% negatively co-expressed genes with HOOK2 (coefficient of -10%)
- consistently the highest negative correlation is observed for gastrointestinal tissues, while the highest positive correlation for whole blood, pancreas and kidney, e.g. for HOOK2 vs. NDE1
there is a known interaction between these proteins, present in protein-protein interaction graphs:
- with at least a few reports about them forming a complex or directly interacting

AnneCarpenter commented 8 months ago

Sounds like it's possible this is a known interaction, although if not, the expression and protein protein information certainly lend strength to the hypothesis they are involved in similar biological processes!

Given you've confirmed the ORF relationship is 'real' I have renewed efforts to get in touch with someone who is expert (emailed msharma mentioned above).

tjetkaARD commented 8 months ago

Correct,

JUMP-ORF:

HOOK2 vs NDE1: anti-correlation, coefficient: -0.704552
HOOK2 vs NDEL1: anti-correlation, coefficient: -0.679421
HOOK2 vs PAFAH1B1: anti-correlation, coefficient: -0.625391
HOOK2, NDE1, NDEL1, PAFAH1B1 do all have the replicable phenotype

JUMP-CRISPR:
HOOK2,NDE1,NDEL1 are not measured in JUMP-CRISPR

Current knowledge:
PPI Knowledge Graph: still, the interaction is strong, but only at the level of co-mentioning in pubmed abstracts, however in multiple organisms
From a bit old review (2018), https://www.nature.com/articles/s41580-018-0004-3, it can be concluded that they are all suspected, but not confirmed, to be involved in the dynein transport. Also, PAFAH1B1 and NDE1/NDEL1 are considered to be involved separately not together.
These two following reports (later than review, of the author you contacted) state that HOOK2 and NDE1/NDEL1 form a complex: https://doi.org/10.1083/jcb.202005184, https://doi.org/10.1083%2Fjcb.201804183.
As I understand the above: currently there are only one source of data to corraborate the relationship we see in JUMP-ORF. As of now, curated databases of interactions (like Biogrid or IntAct) do not accept it as confirmed. Morphological data gives therefore strong independent evidence.
Importantly, dynein is a major microtubule-based motor - so a process that should be well visible in one of cell painting channels.

AnneCarpenter commented 8 months ago

Negative response from msharma: "Thank you for your email. I am excited to learn about your findings on Hook2 and that the morphological changes on Hook2 over expression are opposite to the dynein regulators like Nde1. It is currently not possible for my lab to do experiments to test some of these predictions, as we haven't worked on Hook2 for the past couple of years. My lab mostly works on lysosome fusion machinery involving small GTP binding proteins and tethering factors. These proteins are Rab7, Arl8b and HOPS complex. In case you do find anything related to these proteins or other lysosomal proteins in your analysis, I will be happy to discuss it further with you.

Best regards,

Mahak"

AnneCarpenter commented 8 months ago

Next steps: I believe from the above that this relationship is not a shock - it's independent evidence but not a dramatic discovery. If that is correct, then perhaps we just ask the researcher above if they'd like to synthesize all this info to write a paragraph for the paper describing this as a validated (somewhat already-known) story? Let's discuss.

AnneCarpenter commented 7 months ago

@tjetkaARD Do you feel comfortable writing up this information into a few-sentence blurb for the paper? Or do you wish someone else to do so? (we could assign to Sam or Jess). It seems we don't need to bother the original researcher though we could ask them to review what we write to be sure it's accurate.

tjetkaARD commented 7 months ago

@AnneCarpenter - yes, will do. 4-5 sentences should be fine. Finalise here or somewhere in the document?

AnneCarpenter commented 7 months ago

Straight in the manuscript document would be great. Here is a link to a section stub for this finding: https://docs.google.com/document/d/160QYCeJXMJMPgvYcirkkKaY99r6dZq6FMqYGE61CfFc/edit#heading=h.a0tdqv1hvqwh

tjetkaARD commented 7 months ago

With the update of Knowledge Graph Methodology the status of discovery has changed.

Initially

Previous KG data indicate not very strong connection between HOOK2/NDE1/NDEL1: source: https://github.com/broadinstitute/2023_12_JUMP_data_only_vignettes/issues/7#issue-2044375145

Updated data

Current KG data indicate quite strong connection between HOOK2/NDE1/NDEL1: source: https://github.com/broadinstitute/2023_12_JUMP_data_only_vignettes/issues/7#issuecomment-1884777462

Nonetheless, I think it is still a solid story to write up given rather recent and scarce literature and focusing more on the mechanism itself, connected with the measured CP channel.

auranic commented 7 months ago

Hi, sorry for this change in scoring. Actually, we made yet another iteration adjusting for the reliability of the function prediction from KG and not simply a correlation.

The updated scores are available here: https://drive.google.com/drive/folders/1QWY8itTMeR3pGIt2kWIS5NiOPLhazg4S?usp=sharing

With this adjustment, the score between HOOK2 and NDE1 becomes a bit lower and can be considered as not strong if we accept 0.8 threshold.

However, I do think this connection scores relatively high because of co-citation of these genes, as reported in STRING

AnneCarpenter commented 7 months ago

Do you still agree the next step is to write this up, Tomasz? The researcher we contacted found it not a shocking result so that fits with there being some evidence out there, but still a nice 'discovery' (somewhere in between a novel discovery and a well-known validation).

tjetkaARD commented 7 months ago

@AnneCarpenter - yes; I will write it up until Monday.

AnneCarpenter commented 7 months ago

No rush, mid-February is quite fine though I'm always excited to see text added to the paper!

AnneCarpenter commented 7 months ago

Here's the draft version! I've sent to msharma in hopes they can confirm it makes sense to an expert.

Screenshot 2024-02-14 at 7 52 16 AM

AnneCarpenter commented 6 months ago

From msharma:

"My apologies for a delayed response. I read the text that you have sent. It is quite interesting that Hook2 has a negative correlation with LIS1 and NdeI. I want to reiterate that I am not an expert on dynein regulation (it is quite complex and outside my expertise!!!), thus, I am unable to fully comprehend the strong anti correlation between these sets of dynein regulators. One suggestion is to see if other activating adaptors of dynein like Hook2 such as BicD2, Hook1 and Hook3, Spindly etc. also show this negative correlation with NdeI and LIS1. The labs of Samara Reck Peterson and Erika Holzbaur have a significant body of work on dynein regulation and on the activating dynein adaptors. "

So I did email Dr Reck Peterson just now. We could make lists (using ORF relationships) of the top ~20 matches to the 4 core genes above to see if any other dynein related proteins arise but will hold off on that for now (unless it's easy to just expand the heatmap to include the nearest ~10 genes of each of the 4 genes?).

AnneCarpenter commented 6 months ago

Confirmed by Sam Reck-Peterson! "It makes sense to me that these proteins would have similar localization! Hook2 is an activating adaptor for dynein (one of several dozen proteins that can activate dynein). NDE1, NDEL1 and PAFAH1B1 (also know as Lis1) are also important for dynein function."

AnneCarpenter commented 6 months ago

This recent paper showed HOOK3 in some results, but HOOK2 doesn't show in their hit list, making our connection still new/interesting: https://rupress.org/jcb/article/223/5/e202306048/276601/Genome-scale-requirements-for-dynein-based

niranjchandrasekaran commented 4 months ago

connections

Here's how the connections look in the new profiles. I also included the other genes suggested by msharma

ORF		BICD2	HOOK1	HOOK2	NDE1	NDEL1	PAFAH1B1
BICD2	1	0.09	0.14	-0.05	-0.04	-0.01	0.12
HOOK1		1	0.56	-0.59	-0.53	-0.47	0.41
HOOK2			1	-0.66	-0.61	-0.61	0.31
NDE1				1	0.91	0.86	-0.3
NDEL1					1	0.91	-0.27
PAFAH1B1						1	-0.23
SPDL1							1

CRISPR None of these connections are present in CRISPR either because they are not in the dataset or they don't have a phenotype.

KG Except for the connection between HOOK1 and HOOK2 none of the other connections of HOOK2 are known. Perhaps HOOK2 is not well studied.

I also looked for all the genes that are anti-similar to the four main genes of interest in this issue to check if there are other dyenin related proteins. But I couldn't find any.

Anti-similar to HOOK2

	HOOK2
NDEL1	-0.614541
ST3GAL4	-0.413381
PLAU	-0.42775
NDE1	-0.6563
PURG	-0.440695
PDILT	-0.414814
PAFAH1B1	-0.610536

Anti-similar to NDE1

	NDE1
NIN	-0.427946
HOOK2	-0.6563
ATCAY	-0.532888
PAFAH1B3	-0.530362
CRACR2B	-0.532522
NEK9	-0.412943
PAFAH1B2	-0.585289
FAM114A1	-0.436914
TGIF2LX	-0.430285
HOOK1	-0.589538

Anti-similar to NDEL1

	NDEL1
NIN	-0.418717
HOOK2	-0.614541
ATCAY	-0.490127
PAFAH1B3	-0.484632
CRACR2B	-0.513532
NEK9	-0.40055
PAFAH1B2	-0.514465
FAM114A1	-0.419338
HOOK1	-0.532928

Anti-similar to PAFAH1B1

	PAFAH1B1
HOOK2	-0.610536
ATCAY	-0.450775
CRACR2B	-0.474904
NEK9	-0.421749
PAFAH1B2	-0.416855
FAM114A1	-0.435566
HOOK1	-0.473261

niranjchandrasekaran commented 3 months ago

Notebook

The heatmap shows the percentile of the cosine similarities (1 → similar, 0 → anti-similar). The text is the maximum of the absolute KG score (gene_mf__go, gene_bp_go, gene_pathway). I set a KG threshold (like we previously had) of 0.4. If connections have a score lesser than this threshold, then the connection is considered to be unknown. The KG scores were downloaded from Google Drive: ORF and CRISPR. The diagonal of the heatmap indicates whether a gene has a phenotype (False could also mean the gene is not present in the dataset).

ORF

ORF-connections-BICD2-HOOK1-HOOK2-NDE1-NDEL1-PAFAH1B1-SPDL1

CRISPR

CRISPR-connections-BICD2-HOOK1-HOOK2-NDE1-NDEL1-PAFAH1B1-SPDL1

tjetkaARD commented 3 months ago

I think it corraborates the paragraph that is present in the manuscript (only numbers needs to be adjusted).

niranjchandrasekaran commented 4 weeks ago

Notebook

This connection is not affected by plate layout. ORF-plate-layout-PAFAH1B1-NDE1-NDEL1-BICD2-HOOK1-HOOK2-SPDL1

AnneCarpenter commented 3 weeks ago

@niranjchandrasekaran can you please confirm May 24 is the latest/correct correlation matrix and KG values for this set? (note to self: it's ok there's no CRISPR correlation matrix because only one gene of this set was tested in CRISPR).

Reading thru all this and the paragraph in the paper, I think this story is 'done' but may need reconciling of final figures with the text that describes them in the paper.

niranjchandrasekaran commented 3 weeks ago

can you please confirm May 24 is the latest/correct correlation matrix and KG values for this set?

Can confirm.

broadinstitute / 2023_12_JUMP_data_only_vignettes