We use single cell protein quantification & single cell FISH to map #spatial interactions in genetic mosaicism & tumor microenvironment in #Glioblastoma!
Here, we aimed to better understand how genetically different GBM populations *spatially* interact with each other & further with the tumor microenvironment
We studied this via highly multiplexed spatial protein profiling & single cell spatial mapping of FISH in selected drivers
To do this, we constructed a tissue microarray (TMA) from 17 formalin-fixed paraffin-embedded (FFPE) IDH wild-type GBM tumors.
From each tumor block, we randomly selected 3–4 cores (61 in total) to investigate patterns of local intra-tumor heterogeneity within each tumor biopsy.
We used 3 slides of the multiregion TMA for single cell profiling of:
- EGFR, CDK4 & PDGFRA DNA by multicolor FISH (35K cells)
- specific-to-allele PCR-FISH (STAR-FISH, 26K cells)
- protein production, NanoString GeoMx digital spatial profiling (DSP, 79 proteins)
Main findings👇
1. Local heterogeneity in GBM microenvironment (GeoMx DSP data)
DSP allows quantification of oligo tags released by UV-illuminated regions of interest (ROI) selected by immunofluorescent staining.
It enables simultaneous antibody staining for 10s of markers & control antibodies
Here, we selected 96 ROIs from the 61 TMA cores, with 2–3 ROIs in cores with heterogeneous immunofluorescent staining.
These ROIs were then subjected to DSP for 79 protein markers associated with GBM biology, neurobiology, and immunology.
We observed interesting spatial patterns:
- protein levels of immune genes (CD68, HLA-DR, CD11b, CD45, IBA1, CD163, CD14) correlated across tumors
- members of the WNT pathway correlated with TCF/LEF pathway
- clusters of immune & neuronal markers were negatively correlated
Since the TMA cores represent spatially distant regions of the same tumor fragment with possibly different compositions, we assessed the variability of protein production between ROIs from distinct tumors relative to the variability within tumors (Kruskal-Wallis H test).
The ratio of inter-to-intra tumor variability was quite different across the profiled proteins.
For example, vascular cell adhesion molecule (VCAM) & CD31, both associated with endothelial function & tumor vascularization, showed high variance between regions of the same tumor.
On the contrary, immune-related proteins s.a. CD45, as well as EGFR and p53 (often overexpressed & mutated in GBM), showed particularly high inter vs. intra-tumor variability
This may indicate that immune state & major drivers are less variable between biopsies in the same tumor
2. Single-cell genetic heterogeneity in GBM
We performed multiplexed FISH for EGFR & CDK4 on the multicore TMA (35K nuclei in 17 tumors).
We chose these 2 genes because their copy number amplifications characterize genetically distinct GBM states (+PDGFRA, but no signal here).
Therefore, each cell was assigned to one of the following 4 genotypes:
- E (EGFR amplified & not CDK4 amplified)
- C (CDK4 amplified & not EGFR amplified)
- EC (EGFR & CDK4 amplified)
- N/O (no amplification of either EGFR or CDK4)
We found variation in the frequency of cells of each genotype within each tumor.
Although many tumors had relatively high E cell frequency, C & EC cells were also present in most.
There was also variation between cores taken from the same biopsy, meaning within small regions.
The relative proportions of the different genotypes followed interesting correlation patterns (each point here represents the ratio of the cells with a particular genotype to the total number of cells within a tumor; beware the axes don't go all the way 0-1).
3. Tumor stratification by co-occurrence of EGFR & CDK4 in same vs. different cells
Our dataset of single cell genotypes allows investigating differences btwn tumors in which these alterations tend to co-occur in same cell & tumors in which they do not(different cells or absent)
For this, we calculated the odds ratio (OR) of detecting co-amplified cells within each of the 17 GBMs.
The OR = the likelihood that a cell with EGFR amplification also harbors a CDK4 amplification, relative to the likelihood of a CDK4 amplification without EGFR amplification.
OR = 1: amplification in one of these 2 genes has no effect on the other being amplified.
OR above/below 1: increased/decreased tendency for EGFR & CDK4 amplifications to co-occur in same cell vs. what is expected if one amplification had no effect on the other one occurring.
Based on OR values, we separated tumors into 3 classes:
- ORlow: decreased tendency for EGFR & CDK4 same-cell co-amplification (OR <<1)
- ORhigh: increased tendency for EGFR & CDK4 same-cell co-amplification (OR >>1)
- no effect (OR ~ 1)
The frequencies of E cells (only EGFR amplification) & N/O cells (neither EGFR nor CDK4) were different between ORhigh & ORlow groups, while C & EC cells were similarly distributed
It can then be that low OR is driven by increased frequency of cells with only EGFR amplification
Both the average & the entropy of EGFR copy number values in E cells was significantly higher in ORlow tumors (unlike CDK4).
This suggests that it may be that EGFR amplifications in E cells in ORlow tumors are qualitatively different from those in E cells in ORhigh tumors.
A possible explanation is that EGFR amplification is generated by extrachromosomal DNA (ecDNA) fragments in ORlow tumors.
ecDNA fragments contain an oncogene & an enhancer element, allowing for increased gene expression.
Their presence has been documented in GBM.
We tested this by running AmpliconArchitect on whole genome seq. data from 3 ORlow & 3 ORhigh tumors.
ORlow tumors indeed showed a complex structure of genomic rearrangements involving EGFR amplicon, versus simpler amplification patterns in ORhigh tumors
The in situ single cell genotyping dataset that we generated contains spatial coordinates for each nucleus recorded in an image.
So, based on this data, we can compute distances between spatial locations of cells with distinct genotypes in ORlow & ORhigh tumors.
Interestingly, we find that:
- E cells in ORlow tumors are closer to each other than ORhigh (same distance for C)
- C cells show a tendency towards tighter clustering in the tissue
This suggests that cells with only CDK4 amplif are less migratory & remain closer after division.
4. The co-occurrence of EGFR & CDK4 amplifications is associated with an immunosuppressive tumor microenvironment
ORhigh tumors were enriched in proteins associated with immune cell infiltration & an immunosuppressive microenvironment (CD163, IBA1, CD14, CD45, CD11b, HLA-DR).
Immunofluorescent staining of the original sections from ORlow & ORhigh tumors confirmed huge differences in infiltration of CD163+ cells.
Interestingly, this can suggest that a higher relative frequency of EGFR & CDK4 co-amplif. is associated w/ a more immunosuppressive state!
Phew, that's it for this long thread!
TL;DR:
- EGFR-CDK4 co-amplification corresponds to distinct alteration and expression trends. This co-amplification is subclonal & can only be recovered from single-cell data (hidden in bulk)
- it is possible that our FISH data identified tumors harboring extrachromosomal DNA EGFR as a distinct class of tumors, associated with low immune marker expression & high levels of EGFR protein
- immune cells infiltrating ORhigh tumors may be predominantly immunosuppressive
- even though preliminary, this suggests that the assessment of co-amplification of EGFR and CDK4 at the single-cell level by FISH could serve as a proxy of immune status (much easier & reliable to assess than standard immunofluorescence for immune infiltration)
I need to raise awareness about an important point in #scRNAseq data analysis, which, in my opinion, is not acknowledged enough:
‼️In practice, most cell type assignment methods will fail on totally novel cell types. Biological/expert curation is necessary!
Here's one example👇
Last year, together with @LabPolyak@harvardmed, we published a study in which we did something totally awesome: we experimentally showed how a TGFBR1 inhibitor drug 💊 prevents breast tumor initiation in two different rat models!
1. Define a generative program consisting of a syntax tree & a set of hierarchical constraints 2. Compile the program in (1) into an energy function 3. Optimize the function via simulated annealing. The solutions are the artificial proteins.
Inspiring Symposium on Cancer Prevention @EACRnews
95% of cancer drugs fail. 94% do not improve life quality.
An ounce of prevention is worth a pound of cure. (B. Franklin)
Cancer prevention is tremendously difficult. But it is also necessary.
We need to shift our focus.
How to move from developing cancer treatments to cancer prevention? @cohen_cyrille
How to change the single gene/ single mutation paradigm for holistic approaches considering multi-omics, lifestyle, exposure and cells as a whole? @AzraRazaMD
How much does the environment matter? Can we prevent cancer by modulating exposure? @CBrisken
Which neoantigens to target? Shared or unique mutations? Overexpressed genes?
Will eliciting immune responses via vaccines help prevent tumors in high-risk populations? @emmyverschuren
If you are already familiar with Graph Neural Networks, but still want to better understand the maths behind in a formalized logical framework, I recommend the following book/paper by @mmbronstein@joanbruna@TacoCohen@PetarV_93