Proud to share our latest work in @nature, co-led by @chadi_elfarran, @CouturierMDPhD & @BradEBernstein! We developed a new framework to understand immunomodulatory myeloid cells in #glioma & lay a foundation to develop more effective immunotherapies🧵 nature.com/articles/s41586-0…
Feb 26, 2025 · 5:54 PM UTC
20
47
230
49,880
Quick tangent: we are recruiting postdocs! This work lays a foundation of discovery & translation that I’ve brought to my own lab @caseccc & @CWRUSOM – where we have built a fun, collaborative, & intelligent team driven to make an impact. Join us! tymillerlab.org/our-team/
1
1
898
6 yrs ago, when I started residency @MGHPathology & joined Brad's lab, I asked a ? :
“What’s our best shot at a cure for #glioma patients?”
A: Immunotherapy.
? : “How do we make immunotherapy effective for glioma?”
A: Start by tackling the immunosuppressive myeloid cells.
1
1
5
691
So that’s what we set out to do. While we knew about malignant cells in glioma from scRNA-seq studies & had some consensus in the field, that’s not the case for myeloid cells. So 1st, we set to create consensus myeloid programs across gliomas(which are diverse = need many tumors)
1
3
534
I could go on for 20 tweets about how we got to these programs (it took 2 years to develop the strategy we felt captured the true consensus programs), but the key was utilizing cNMF (from @DKotliar @PardisSabeti) rather than the standard Louvain clustering w/UMAP (see methods).
1
3
507
There were 14 superimposable programs. 5 cell identity programs (defines cell type of cell), and 9 activity programs (what cell is doing). Notably, 4 of these activity programs seem immunomodulatory in their activity, with 2 inflammatory and 2 suppressive.
1
1
447
Surprisingly, these 4 immunomodulatory were the most utilized programs across myeloid cells (d). Nearly all (91%) of myeloid cells in our gliomas expressed at least one of these 4 programs.
1
427
And they were shared across myeloid cell types = the same activity program was being expressed in different microglia, macrophage, and monocytes (and even cDCs).
1
418
This made us rethink how we should view myeloid cells. Rather than taking a cell type-centric approach, we decided to focus on the immunomodulatory programs and study myeloid cells in that way (in a cell-type agnostic way). This plot becomes very useful to do that.
1
388
If we think the immunomodulatory activity programs may be the most important programs contributing to the immune state of the tumor, then what drives these programs? It wasn't cell type. Also, it's not the cell's origin(we did lineage inference w/MAESTER): nature.com/articles/s41587-0…
1
1
375
Side note, we found that myeloid cells were so plastic that even some cells expressing microglia programs were derived from circulating monocytes. We validated this ex vivo - when we applied peripheral monocytes to patient organoids, they infiltrate & express microglia markers!?
1
2
446
We also looked to see if IDH mutation was a driver of myeloid activity. While the Microglial inflammatory program was very enriched, and the 2 immunosuppressive programs were depleted, this turns out to be entirely driven by tumor GRADE not IDH mutation.
1
419
Grade = microenvironment under a microscope, so we looked to see what specific tumor environments were associated w/myeloid activities programs using 10X Visium data from @MILOLab2. Our Scavenger Suppressive program was only in hypoxia. The Complement Suppressive? Everywhere else
1
1
441
Using a new spatial regression model by @CouturierMDPhD & @davidsebfischer, we aggregated the spatial association data of tumor niches & our cellular programs across sections. We created a spatial map from the data - each activity program had a distinct tumor niche.
2
2
523
Importantly, do these programs matter clinically? Fortunately, a paper was published recently with a scRNA-seq dataset of 12 patients w/ neoadjuvant PD1 blockade, categorized as responder or non-responder.
1
1
440
The authors found SIGLEC9 is a mediator of resistance. We plotted their cells according to our program usage & then labeled SIGLEC9+ cells. They were quite diverse in our program expression. However, responder/non-responder tumor cells almost perfectly segregated by our programs!
1
424
It turns out that it was the Scavenger Suppressive program, not any cell type, that was associated with immunotherapy resistance, higher Tregs in tumors, and overall worse survival in patients – only discoverable because we can now study this program in isolation:
1
1
436
Given their apparent importance, we dove in to discover the mechanisms underlying these immunosuppressive programs. We used snATAC-seq data to identify transcription factors at the heart of these programs and then identified upstream regulators of these TFs.
1
425
We identified IL1B as a major driver of the Scavenger suppressive program, which is being produced by monocytes that enter the tumor expressing the System Inflammatory program. Likely an evolutionary feedback loop to reduce inflammation in the brain! We confirmed this w/organoids
1
2
437
Using clinical data, we found the Complement Suppressive program is specifically & irreversibly driven by dexamethasone, a potent corticosteroid given to most patients for symptom & surgical management. High IFNg only partially rescued. Most IO trials in GBM allow some dex use...
1
446
Excitingly, we could reprogram the immunosuppressive macrophages back to their default inflammatory state using a p300 inhibitor from @genentech
1
1
466
These 4 immunomodulatory programs dictate the overall immune state tumor and are very interconnected!
1
1
9
875
Taking all this data together, and through many conversations with a lot of really smart people over the last 5+ years, we now think about myeloid cells in glioma through this framework (which is probably applicable to other solid tumors):
1
2
310
Notably, we would not have been able to find these important programs and study them in isolation with a traditional clustering/UMAP strategy. Clustering is good for finding cell types, but not activities, which are shared across clusters.
1
1
312
The clusters in prior important studies of myeloid cells in glioma (which taught us a lot) end up being composites of our programs – with our cell identity programs being captured well by 1-3 clusters, but the activities programs being shared across many clusters.
1
309
In summary, we utilized a new analytical tool, cNMF, to define consensus myeloid programs in glioma & revealed 4 dominant immunomodulatory programs that were previously hidden. Being able to study them in isolation reveals their drivers and ways to target them.
ALT Microglial Inflam.↔️low grade, only in brain Systemic Inflam. = default inflam response, seen systemically and in all tumors Scav. Supp.↔️hypoxia, IL1B, TGFB; only in brain tumors Complement Supp. <-Dex, endogenous stress hormones; seen in all tumor types
1
1
1
309
Our goal is to bring some consensus to the field & make this a resource so that we can work on solving this problem together. Towards that, we’ve
1) created a GitHub where we share all of our code needed to analyze your own datasets within this framework: github.com/BernsteinLab/Myel…
1
1
241
2) deposited all the data and created a portal to explore your favorite genes within these datasets through the @broadinstitute's single cell portal. I suggest using the "Quadrant plot" clustering option
singlecell.broadinstitute.or…
1
1
229
3) have created an online tool where you can upload a gene expression matrix of your myeloid cells and it outputs program usages of these consensus programs for all your myeloid cells.
consensus-myeloid-program-ca…
1
1
222
Great team effort with @chadi_elfarran, @CouturierMDPhD,@ZeyuCHEN19,Josh D'Antonio,Julia Verga,@mav_tweets,Nicolas Gonzalex Casto,Evelyn Tong, @tariq_dh ,Andrew Chiocca,@davidsebfischer, @MILOLab2, @JennGuerriero,Kevin Petrecca,@MarioSuva,@shaleklab,@BradEBernstein!!
1
1
2
503
A huge thank you to the patients & families that contributed to this study, the orgs that help fund this work - @BrainTumourOrg (Future Leaders program), @theABTA,@theNCI,@NIH,@CIHR_IRSC & our supportive institutions @MGHPathology @DFCI_CancerBio @broadinstitute @MIT @mcgillu!
2
3
347
