Mockup 08 · traverse & request · push results to a shared web interface

A grad student traverses to the threshold, then sends a Request back

From Sean's Evidence, the reproducer follows the lineage down to the segmentation choice, resolves the git/data pointers across system boundaries and watches the walkthrough, hits the gated raw stacks gracefully — then opens the reverse current and sends a Request for a follow-up (R2, R4, R5, R7, plus the first-class Request: results out, requests back).

🔒 mira.science/n/sg-composition-019ea6d6 ⌘ L
MIRA · shared dashboard
Stress-granule formation · Vogel × Kate
M viewing as Kate's student
raw stacks: request access
Evidence · supports a Claim · grounded in a Study

Stress granules change composition over time

Across both stresses, the correlation between the two granule proteins climbs from negative to positive over ~10–15 min — they first exclude each other, then co-localize as the granule matures. Measured per granule by live Lattice-SIM imaging.

S Sean · Vogel Lab · shared with Kate's lab · node sg-composition-019ea6d6
+1 0 −1 Pearson r · Protein 1 · Protein 2 stress 30 min → time since stress (min) ≈10–15 min arsenite osmotic

What's in the dish

Cell lineHeLa
Stressarsenite · osmotic
ChannelsP1–GFP · P2–mKate
ImagingLattice SIM · Elyra 7
Readoutco-localization
Binnedby time since stress
arsenite osmotic
illustrative · unpublished — workshop prototype
M REPRODUCER
“How exactly did Sean segment the granules?”
Because the segmentation choice can change all the measurements — so before building on this, follow the lineage down to the method that produced the curve.
traverse: Evidence ← grounds ← Study → follows → Protocol

Follow the lineage to the method

Each card is one discourse node. Typed edges connect the result to the experiment and the protocol — pointers only; no raw payloads travel over the web, and they resolve across system boundaries (vault → GitHub → the lab array). The rail draws down to the payoff: the segmentation choice. Every node also carries the author’s own words — open Sean’s notes or the full protocol on a node to read, unedited, the human-written grounding each summary is condensed from.

Evidence · this node you are here
Stress granules change composition over time
Across both stresses, the correlation between the two proteins goes from negative to positive over ~10–15 min — they start by excluding each other, then co-localize — the curve above.
datagranule_composition_by_time.csv (this figure)
Sean's noteshis read on what the curve actually means
  • In both conditions, the Pearson correlation coefficient between the two channels starts negative and becomes positive over time. This implies that the two tagged proteins (Protein 1 and Protein 2) are co-localizing more over time and they even start to coalesce independently of each other in the beginning.
  • While this shows that the proteins start off excluding each other and then mix over time, it is not clear what the nature of this exclusion is. They could be forming different droplets which merge over time, or they could both be forming in the same droplet, but occupying different sub-compartments in the droplet. Another experiment will need to be done to understand the nature/structure of these early SGs.
  • While the two stress conditions look similar after 10–15 minutes, there is a significant difference in them prior to that point. The interpretation of this would depend on the result of the previous point. Although, it does imply that there is a measurable difference in the co-localization of these proteins during different stress, perhaps implying that they are related to how the SGs form in response to a particular type of stress.
Verbatim · RES node · protein names redacted
1 of Sean's working notes withheld — present in his vault, not part of this share.
grounds ↑ Study → Evidence
Study · "the experiment"
Live Lattice-SIM imaging of granule formation at staged times
Protein 1–GFP (endogenous) and Protein 2–mKate (transfected) in HeLa, imaged on a ZEISS Elyra 7 in Lattice SIM on fresh fields of view up to 30 min after stress.
Experiment notesSean's lab-notebook progress, design choices & hypothesis
Progress & notes

2025-08-23 — Kate's student visited Montreal and they sent their cells ahead of time. Khalid is maintaining the cell culture using their protocol. (See linked protocol node.) This cell line has Protein 1 tagged with GFP and we will transfect it with a plasmid expressing Protein 2-mKate. Applied the stress to the cells and imaged them on the microscope to measure the rate/amount/coordination of the two proteins forming stress granules.

2025-10-14 — I have developed the segmentation and subsequent image analysis pipeline. This has yielded initial results (attached). There were several design choices which had to be made including:

  • What channel to use for the segmentation, or whether to use the SUM or the MAX (pixel-wise) for the segmentation.
  • What intensity threshold to use to exclude local maxima which are due to background noise? (This is a parameter of the ijm macro and was set by choosing a value that is just above the background max intensity.)
Hypothesis

We have multiple hypotheses that are addressed by this experiment:

  1. Stress granule components begin to form droplets independently of each other, and these components later mix into the same droplet to give the SGs we typically observe.
  2. This process will be perturbed when we change the type of stress because different types of stress have different sensors which trigger the SG formation mechanism. The different nature of the formation (with respect to the two tagged proteins) will give us insight into how these two proteins are related to the downstream effects of the stress sensing mechanism which lead to the droplet formation.
Verbatim · EXP node · protein names redacted
rawNDTiff z-stacks ≈ 80 GB · Elyra 7request access
video 2-min walkthrough 2:04 · Loom
2-min walkthrough — no Zoom call, re-watchable in 3 years.
follows ↓ Study → Protocol · the payoff
Protocol the method that matters
HeLa culture → stress → Lattice-SIM → segment on per-pixel SUM → composition over time
The full pipeline behind the curve. The load-bearing step is the segmentation — how a granule is defined sets every co-localization number that follows.
Segmentation choice what the grad student came for
Segment onper-pixel SUM of both
Detector3D Suite · maxima + FWHM
Thresholdjust above background
Readoutregionprops, per channel
gitvogel-lab/sg-imaging · segment_stress_granules.ijm
The full protocolevery step & parameter, as written at the bench

The pipeline summary above is condensed from three protocol nodes the Study follows. Their verbatim text:

① Cell culture

TODO: Get growth information from Khalid and Kate's students.

② High-resolution stress-granule imaging (Lattice SIM)
  1. Cultured cells were adhered to a coverslip and placed in a well containing 1µl of the media they were grown in. This cell line was tagged with endogenous Protein 1-GFP and was transfected with a plasmid expressing Protein 2-mKate.
  2. As quickly as possible:
    1. The media was removed from the well with a pipette.
    2. New media containing the stress (Sodium Arsenite or osmotic stress) was added using a pipette. Do not pipette directly onto the coverslip. Deposit the new media on the side of the well to avoid disturbing the cells or applying any mechanical stress.
    3. Start a stopwatch to calculate the amount of time since the stress was applied.
  3. Place the cells on the Elyra 7 stage and shut the enclosure in preparation for imaging. The time from the exchange of the media to the first image acquisition should be as short as possible to observe the stress response in its early stages.
  4. Scan over the cells using the bright-field to identify cells which are healthy and which are close enough together that you can get a good amount (purely to increase the data acquisition efficiency). You can position the center of the z-stack in the bright-field to avoid using the laser and initiating any photobleaching so that we can quantitatively compare the intensities in the analysis step.
  5. Define the "center" of the z-stack in the ZEN software controlling the microscope.
  6. Switch to the Lattice SIM acquisition track and acquire a z-stack using the parameters outlined below. Note the time since the stress was applied using your stopwatch and include this time in the filename for the raw image.
  7. Move to a NEW field of view so we have fresh cells which have never been exposed to the laser and repeat the imaging steps above (4–6). Repeat this until you have imaged for 30 minutes after the stress application.
  8. Repeat steps 1–7 with a new coverslip in a new well of cultured cells to generate replicates.
  9. Process the RAW image files using batch mode in ZEN to perform the SIM deconvolution and generate the super-resolution images. Use the parameters in the table below. Use the default parameters otherwise.
Imaging parameterValue
Number of Z slices25
Z slice spacing300 nm
488 nm laser power0.8
561 nm laser power0.8
Exposure time30 ms
Number of SIM phases13
Use Z-piezoYes
Processing parameterValue
SIM SquaredNo
Leap Mode3D Leap
Processing StrengthWeak
Scale To RawYes
Gaussian Fit MethodFast Fit
③ Segmentation & composition measurement
  1. Crop individual cells into their own 3D stack tif images using Fiji from the processed SIM data.
  2. For each tif of a cell, run the segment_stress_granules.ijm ImageJ macro to generate a label image which segments the stress granules in the fluorescent image. This is done using the 3D Suite plugin to identify local signal maxima and then fit a 3D gaussian to use a FWHM threshold for each object (stress granule). Because there are two signal channels, this was done on the per-pixel SUM of the channels to prevent bias toward segmenting objects which have more or less of a single component.
  3. Then in python (see python script in data), the fluorescent tif is loaded and the label image tif is loaded, and the regionprops function is used to compute the statistics for each color channel on each spot.

Analysis & statistics. Each cell was binned into time bins based on how long it has been since the stress was applied.

Verbatim · 3 × PRO nodes · protein names redacted
Why this lineage matters. A bug in segmentation, found 3 months later? Everything downstream — this result, anything built on it — can be invalidated and recomputed, because the method is an addressable node, not a buried detail.

Discuss the result — separate from the method, so nothing gets buried

💬 Discussion (3) — kept separate from the result & method
K
Kate · 2d
Why segment on the SUM of the two channels, and not each one separately?
S
Sean · 1d
Per-channel biases toward whichever protein is brighter — the SUM keeps a granule a granule even while one component is still excluded.
K
Kate · 4h
Makes sense. Can you push the time-course below 5 min? The early gap between the two stresses is the interesting part.
Viewing only — sign in to comment
Read-only · pointers resolve across system boundaries · rendered from MIRA JSON-LD over KOI
vault → GitHub → 25 TB array