Poster - Supplementary Information

Conference Poster Autumn 2026

I am presenting a poster on my recent work at two upcoming conferences: Cambridge Lymphoma Biology International Symposium (14-15th September 2026) and the 7th NF-kB International Congress in Barcelona (6-9th October 2026). Travel bursaries to attend these events have been kindly provided by the UK CLL Forum and by Brighton and Sussex Medical School.

Supplementary Methods

Experimental Methods

Spectral flow cytometry data was collected using the panel below and fingerprints were generated using Julia 1.8.5 as described by Jayawant et al., 2023 and Vareli et al., 2025 (see Publications for further details).
Nuclear ELISAs were performed using ActiveMotif TransAm NF-κB Family Kit, using nuclear extracts prepared with the ActiveMotif Nuclear Extract Kit.
Fractionation was validated using tubulin and lamin western blots (not shown).
RNA sequencing was performed by ActiveMotif. QC analysis was performed in-house and is shown below. All analysis was performed using R 4.6.0.

Computational Methods

Ordinary differential equations were solved using Julia 1.8.5, to calculate the relative abundances of NF-κB subunits.
Parameter optimisation was performed using a combined simulated annealing and particle swarm optimisation (PSO) approach.
Initially, a 39-parameter space was explored with bounds of 0.01 and 10 (10 conditions; three NF-κB synthesis parameters per condition; plus nine global coupling parameters between NF-κB and BCL2 family members).
This provided a good fit, but in order to further refine the model a second round of PSO was performed to optimise MCL1 degradation (with the 39 "best" parameters from the previous optimisation fixed).
This provided an excellent fit with minimal error between the z-scores from the experimental data and those from the simulated cells.
1000 cells were simulated per conditions (10000 simulations in total).

Figure 1

Schematic representing methods. We used a systems biology model containing molecular detail about NF-κB subunits, coupled to upstream TLR and BCR signalling. To this model, we added regulatory links between different NF-κB subunits and BCL2 family members (initially with arbitrary rates).

Experimental spectral flow cytometry was collected for surface markers, three NF-κB subunits (RelA, cRel and RelB), and three BCL2 family members (BCL2, BCLXL and MCL1). Relative protein abundances were calculated (fingerprinting approach previously described by Jayawant et al., 2023). NF-κB activity was also assessed using nuclear ELISA assays for RelA, cRel and RelB. Bulk RNA seq provided data on transcriptional activation. All data was collected for two co-culture conditions: 24 hour culture with non-transfected (NTL)-NIH-3T3 cells (control condition), or with NIH-3T3 cells engineered to express human CD40L (to mimic the interaction with T-helper cells in the tumour microenvironment).

The flow cytometric data (fingerprints) were used to inform the systems biology model. Initially, we attempted to manually fit the model to the experimental data, however this was not successful. Instead, we opted for a machine learning approach. Using a combination of simulated annealing and particle swarm optimisation, we were able to successfully reduce the error between the simulations and the target experimental data.

Figure 2

(A) Fold change of protein abundances in all samples following 24 hour co-culture with hCD40L-expressing NIH-3T3 fibroblasts (compared to co-culture with non-transfected NIH-3T3s), as measured by spectral flow cytometry. Data for RelA, RelB, cRel, BCL2, BCLXL and MCL1 is shown. Type of sample (CLL primary, CLL cell line, DLBCL cell line, RS cell line) is indicated with shape and colour. n = 18; ** = p < 0.01, *** = p < 0.001, **** = p < 0.0001; one-way ANOVA with multiple comparisons. A subset of patient data presented here w as acquired by Dr Kinga Pénzes.

(B) Volcano plot representing CD40L vs NTL conditions for 6 primary CLL patient samples and U-RT1. CD 40L gene omitted. Orange = increased gene expression following CD40L co-culture, blue = increased gene expression following NTL co-culture, grey = not significant. Dashed lines represent

significance threshold.

(C) Nuclear abundances of NF-κB family members in primary CLL samples following 24 hour co-culture with NTL- or CD40L-fibroblasts. Nuclear extracts were prepared using ActiveMotif Nuclear Extract Kit and fractionation was validated using tubulin and lamin western blots. Nuclear ELISAs were performed using ActiveMotif TransAM NF-κB Family Kit. n = 6; * = p < 0.05, ** = p < 0.01; paired t-test.

Figure 3

(A) Violin plots showing best 25 iterations of parameter optimisation within bounds of 0.01 and 10 for global coupling parameters between NF-κB and BCL2 family members, across 10 distinct conditions (six primary CLL samples, one CLL cell line, two DLBCL cell lines and one RS cell line). Tight violins indicate that the model is not overfitting. In order for the simulations to match the target experimental data, the influence of RelA on BCL2 family proteins is decreased (indicated by low parameter value), while cRel has a large regulatory role. In the optimised model, BCLXL is also regulated by RelB.

(B) Heat maps showing target experimental data (z-scores; top) and results for a single optimised simulated cell (bottom). Parameters were optimised to match the experimental data in a 39-parameter space, with nine global parameters (coupling between NF-κB and BCL2 family members as indicated in (A)), and three specific NF-κB expression rates (RelA, cRel and RelB) per condition.

(C) Experimentally-measured NF-κB/BCL2 fingerprinting based on BCLXL and RelB abundance, and comparison to simulated cell populations. Cell density is indicated with a contour plot and each cell population is shown in distinct colours. The mean of population-specific computational simulations based on the parameter optimisation are indicated with a star, with colours matching their corresponding experimental contour plot. 1000 cells were simulated in each cell population (10000 simulations in total). Violin plots show a two-dimensional representation of the population scale experimental (solid lines) and simulation (dashed lines) data.

(B & C) MEC-1 = CLL cell line, RIVA = ABC-DLBCL cell line, SUDHL-8 = GC-DLBCL cell line, U-RT1 = RS cell line, all others = CLL patient samples.

Figure 4

Schematic representing future work. Our focus in the future involves using our optimised models to identify patient- and condition-specific therapeutic vulnerabilities, which we will test experimentally with specific drugs based on predictions from the model.