Abstract
Background
Established genetic biomarkers in chronic lymphocytic leukemia (CLL)
have been useful in predicting response to chemoimmunotherapy but are
less predictive of response to targeted therapies. With several such
targeted therapies now approved for CLL, identifying novel, non-genetic
predictive biomarkers of response may help to select the optimal
therapy for individual patients.
Methods
We coupled data from a functional precision medicine technique called
BH3-profiling, which assesses cellular cytochrome c loss levels as
indicators for survival dependence on anti-apoptotic proteins, with
multi-omics data consisting of targeted and whole-exome sequencing,
genome-wide DNA methylation profiles, RNA-sequencing, protein and
functional analyses, to identify biomarkers for treatment response in
CLL patients.
Results
We initially studied 73 CLL patients from a discovery cohort. We found
that greater dependence on the anti-apoptotic BCL-2 protein was
associated with prognostically favorable genetic biomarkers.
Furthermore, BCL-2 dependence was strongly associated with gene
expression patterns and signaling pathways that suggest a more targeted
drug-sensitive milieu and was predictive of drug responses. We
subsequently demonstrated that these associations were causal in cell
lines and additional CLL patient samples. To validate the findings from
our discovery cohort and in vitro studies, we utilized primary CLL
cells from 54 additional patients treated on a prospective, phase-2
clinical trial of the BTK inhibitor ibrutinib given in combination with
chemoimmunotherapy (fludarabine, cyclophosphamide, rituximab) and
confirmed in this independent dataset that higher BCL-2 dependence
predicted favorable clinical response, independent of the genetic
background of the CLL cells.
Conclusion
We comprehensively defined BCL-2 dependence as a potential functional
and predictive biomarker of treatment response in CLL, underscoring the
importance of characterizing apoptotic signaling in CLL to stratify
patients beyond genetic markers and identifying novel combinations to
exploit BCL-2 dependence therapeutically. Our approach has the
potential to help optimize targeted therapy combinations for CLL
patients.
Graphical Abstract
[56]graphic file with name 12943_2025_2260_Figa_HTML.jpg
Supplementary Information
The online version contains supplementary material available at
10.1186/s12943-025-02260-7.
Introduction
Over the last decade, targeted agents have transformed the therapeutic
landscape in chronic lymphocytic leukemia (CLL). Although predictive of
response to chemoimmunotherapy, established genetic biomarkers such as
deletion-17p (del(17p)), mutation of tumor protein-53 (TP53), and the
mutational status of immunoglobulin heavy chain variable gene (IGHV)
are less useful in predicting response to targeted agents [[57]1]. For
example, patients with CLL who are treated with continuous Bruton
Tyrosine Kinase inhibitors (BTKi) have similar median progression-free
survival (PFS) irrespective of IGHV mutational status [[58]1–[59]3].
Hence, additional biomarkers that predict response to targeted therapy
would be useful in CLL to rationally guide therapy selection and
optimize outcomes for individual patients.
Several important studies have dissected the diverse array of genomic
aberrations observed in CLL [[60]4, [61]5]. These rich datasets help us
understand CLL biology; however, they do not fully explain the
heterogeneity in response observed with targeted agents. Functional
precision medicine may provide an additional layer of information to
complement genomics, but until recently has not been widely studied in
CLL. Initial work in this area utilized ex vivo sensitivity testing
alongside genome, transcriptome, and DNA methylome analyses as a
drug-perturbation strategy to classify CLL in phenotypic subgroups with
characteristic dependences on the B-cell receptor (BCR) pathway,
mammalian target of Rapamycin (mTOR), and mitogen-activated protein
kinase kinase (MEK) [[62]6]. Subsequent efforts using ex vivo profiling
of response to individual drugs or combinations revealed synergistic
drug effects and genetic dependences in CLL [[63]7] as well as guided
treatment with improved clinical benefits in certain patients with
aggressive hematologic malignancies [[64]8]. Despite the helpful
insights gained from these studies, they did not specifically provide
functional characterization of the intrinsic pathway of mitochondrial
apoptosis, known to be a fundamental aspect of CLL pathophysiology
[[65]9–[66]11].
The mitochondrial apoptotic pathway is regulated by the complex
interactions between pro- and anti-apoptotic proteins of the B-cell
leukemia/lymphoma-2 (BCL-2) family. The interactions of these proteins
determine cell fate by governing the threshold of mitochondrial
outer-membrane permeabilization and cytochrome c (CytC) release,
followed by caspase-dependent apoptosis. BH3-profiling is a functional
assay developed to interrogate the interactions of BCL-2 family members
and thereby measure the proximity of a cell to the apoptotic threshold
(known as “priming”), as well as to identify specific anti-apoptotic
proteins a cell depends on for survival [[67]12, [68]13].
To more completely characterize functional BCL-2 family dependence in
CLL and begin to unravel the relationship of apoptotic priming to
common genomic aberrations and treatment response in this disease, we
studied primary samples from 127 patients with CLL using BH3-profiling.
This included an initial discovery cohort of 73 patients with available
genomic data including targeted-sequencing (targeted-Seq), whole-exome
sequencing (WES), genome-wide DNA methylation profiles and
RNA-sequencing (RNA-Seq). Our findings suggest that dependence on the
anti-apoptotic protein BCL-2 for survival is a functional predictive
biomarker for treatment response, which we further validated
experimentally in cell lines and ex vivo CLL patient samples. Using a
validation cohort of 54 additional primary samples from CLL patients
treated uniformly in a phase-2 clinical trial of the Bruton Tyrosine
Kinase (BTK) inhibitor ibrutinib with the chemotherapy regimen
fludarabine, cyclophosphamide, and rituximab (FCR) [[69]14], we
confirmed our findings in vivo in patients to demonstrate the potential
clinical applicability of this approach.
Methods
Patient samples isolation and usage
Peripheral blood mononuclear cells (PBMCs) were collected and analyzed
from 73 patients of our discovery cohort, and 54 samples of our
validation cohort, with all patients fulfilling diagnostic and
immunophenotypic criteria for CLL. Cells were isolated by density
gradient centrifugation by Ficoll-Paque (GE Healthcare), viably frozen
in FBS (Gibco) supplemented with 10% DMSO. Samples with > 60% viability
and > 85% CD19^+ CD5^+ cells (PBMCs) were used for analysis.
Baseline characteristics of the discovery cohort are in supplementary
material 2 and cited here [[70]6]. For validation cohort, samples were
collected from previously untreated patients in an
investigator-initiated trial of CLL for frontline therapy with
ibrutinib + FCR (iFCR) at pre-treatment and after one-week of ibrutinib
monotherapy (just prior to the first cycle of FCR). Baseline patient
characteristics, study design, and full clinical results were
previously reported [[71]14]. Samples from both discovery and
validation cohorts were BH3-profiled. Detailed BH3-profiling technique
is in the supplemental information.
Multi-omics profiling
Multi-omics profiling, including WES, targeted-Seq, DNA methylation
profiling and RNA-Seq, were previously performed on the same set of
patient samples from the discovery cohort [[72]6]. Clinical outcomes of
these samples were recorded. Data are available in the R data package
BloodCancerMultiOmics2017 from the Bioconductor project
([73]http://bioconductor.org).
Integrative data analysis
Analyses were performed using R3.4, which include univariate
association tests, multivariate regression with and without lasso
penalization, Cox regression, generalized linear models, principal
component analysis, and gene set enrichment analysis. For association
tests between bioenergetic features and genetic variants (i.e. copy
number variants and gene mutations), only those with five or more
variant cases were included. All P-values from association tests were
adjusted for multiple testing by applying the Benjamini–Hochberg
procedure to control false discovery rate (FDR). The full analysis
walkthrough and scripts as Rmarkdown files are available at
(GitHub:[74]https://github.com/Lu-Group-UKHD/CLL_BH3_Profiling).
Statistical analysis
Dotted graph bars depict mean ± SD. Significance tests were displayed
in figure legends together with the number of replicates for each
experiment. Significant P-value was set at < 0.05. Asterisks on graph
bars indicate *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, or
otherwise specified in the figures.
A more comprehensive method section could be found in the supplemental
information file.
Results
Overall apoptotic priming and functional dependence correlate with distinct
molecular features
As the heterogeneity of ex vivo drug responses in CLL has been found to
be partly explained by the genomic background of patient samples
[[75]7], we first evaluated whether differences in mitochondrial
priming and functional dependence on BCL-2 family anti-apoptotic
proteins could complement genomic data to better understand this
variation. We performed BH3-profiling on primary CLL cells derived from
the peripheral blood of 73 patients in our discovery cohort, using 7
different BH3 peptides and mimetic drug in multiple concentrations each
to capture the full range of different anti-apoptotic protein
dependences (Fig. [76]1A, S1). These 7 BH3 peptides and mimetic drug
have previously been characterized and tested to ensure their
specificity to binding and inhibiting their respective anti-apoptotic
proteins to elucidate survival dependences [[77]15, [78]16]. Several
concentrations of peptides and mimetic were included to ensure
sufficient coverage of their effects and reliability in inducing CytC
loss across a dynamic range. Cells were first incubated with individual
BH3 peptides and BH3 mimetic drug that recognize specific
anti-apoptotic proteins to induce CytC release. Greater CytC release
indicates higher dependence(s) on specific anti-apoptotic protein(s),
depicted by the heatmap (Fig. [79]1A, right panel). Significant
heterogeneity was observed, with CLL cells displaying a wide range of
baseline priming and marked variation in dependence on different
anti-apoptotic BCL-2 family members (Fig. [80]1B, S1). Nonetheless, CLL
cells have a predominant pattern of high BCL-2 dependence as shown by
the significant CytC release even at low concentrations of ABT199 (VEN,
BCL-2 specific inhibitor) or BAD (BCL-2 and BCL-xL-specific peptide),
as compared to HRKy (BCL-xL-specific peptide), MS1 (MCL-1-specific
peptide) or FS1 (BFL-1-specific peptide). The results from BAD peptide
here specifically reflect BCL-2 dependence due to the low
HRKy-dependent CytC release and lack of correlation to HRKy (BCL-xL
dependence) (Fig. S2-S3).
Fig. 1.
[81]Fig. 1
[82]Open in a new tab
BH3-profiling overview and its correlations with clinical and genetic
backgrounds in primary CLL cells. A BH3-profiling is a technique used
to measure how close a cell is to the threshold of apoptosis, and to
identify how dependent a cell is on certain BCL-2 family anti-apoptotic
protein(s) for survival. Measurements by flow cytometry are determined
based on the level of cytochrome c (CytC) release, induced by specific
BH3 peptides or mimetic drug, whereby the higher the CytC is released
or loss (%), the higher the dependence(s) is towards its respective
anti-apoptotic protein(s). The heatmap shows peptides or mimetic drug
that measures the overall apoptotic priming of cells and the specific
anti-apoptotic dependences of cells. Diagrams were created either in
BioRender. Chamberlain, S. (2025) [83]https://BioRender.com/y06s314
(left panel) or Powerpoint software (heatmap). B Measurement of CytC
release (%) of 73 CLL patient samples from the discovery cohort,
induced by the BH3 peptides or mimetic. C Heatmap showing BH3-profiling
pattern with heterogenic clinical and genetic background. The CytC
values (area under curve) were centered by mean and scaled by standard
deviation column-wise in order to reveal the pattern among patient
samples with heterogenic clinical and genetic background. D The PCA
biplot shows CLL samples (points) and BH3 peptides or mimetics (arrows)
on the first two principal components (PC1 and PC2). Points represent
individual CLL samples, with distances indicating similarities, while
arrows show variable (BH3 peptides or mimetic) contributions, with
direction and length indicating correlations and influence (strength of
correlations) on the components. Closer points suggest similar samples
and angles between arrows indicate variable correlations. E Boxplots
showing the significant associations (nominal P value < 0.05, Student’s
t-test) between principal components and patient genetic background.
Tri12 – Trisomy 12, WT – wild type, M – mutated, U – unmutated, HP –
highly programmed, IP – intermediately programmed or LP – lowly
programmed CLL based on methylation clusters, Mut – mutated.
F Significant correlations (nominal P value < 0.05, Student’s t-test)
between principal components and the % baseline viabilities of the
cells from CLL patient samples (% live cells) after 48-h culture in
media
To assess the molecular determinants of anti-apoptotic protein
dependence and mitochondrial priming, we integrated our BH3-profiling
data with targeted-Seq, WES, DNA methylation profiles and RNA-Seq data
that were previously reported for this same patient cohort [[84]6]. To
characterize the overall landscape in an unbiased fashion, we first
performed unsupervised hierarchical clustering of all pro-apoptotic
peptides and CLL samples based on the BH3-profiling data (Fig. [85]1C).
This clustering did not reveal distinct groups significantly related to
genomic or clinical features but showed a continuous gradient that
reflects a response pattern of CLL samples to pro-apoptotic peptides.
This common trend was also observed in the principal component analysis
(PCA), where the first principal component (PC1), which explains 60.8%
of the variance in the BH3 profile, was negatively associated with the
priming of all BH3 peptides and thus indicated a more resistant
phenotype (Fig. [86]1D). By correlating PC1 with individual genomic
backgrounds, we observed that CLL cells with trisomy 12 show
significantly lower PC1 values, indicating that these samples are more
responsive to BH3 peptide-induced CytC release and potentially to
drug-induced mitochondrial apoptosis (Fig. [87]1D, E). This is in line
with our previous observation that samples with trisomy 12 show higher
ex vivo drug sensitivity to multiple kinase inhibitors [[88]6].
In contrast, PC2, which explains 21.1% of the variance (Fig. [89]1D),
distinguished the populations of CLL samples that fall under ABT199,
BAD, and PUMA peptides, which indicate higher CytC release and
apoptotic priming (Fig. [90]1B), or HRKy, FS1, MS1, which indicate
relatively lower CytC release and hence more apoptotic-resistant
(Fig. [91]1B). PC2 shows significant correlations with IGHV mutational
status, and with the epigenetic sub-types of CLL (methylation clusters)
(Fig. [92]1E). Mutated IGHV status correlating with negative PC2 values
suggest that these cells are more responsive to ABT199, BAD or PUMA
(Fig. [93]1E), corresponding to stronger CytC release with these BH3
mimetic/peptides (Fig. [94]1B). This is consistent with the more
indolent nature of the disease in CLL patients with mutated IGHV
[[95]17]. In contrast, lowly programmed (LP) CLL or hypomethylation
cluster correlated with higher PC2 values, suggesting that this cluster
is less primed for apoptosis, which corresponds to our recent
publication that CLL cells with high proliferative drive possess
hypomethylation profile and worst outcome [[96]18, [97]19]. PC2 also
showed significant correlation with NOTCH1 mutational status, whereby
wild-type NOTCH1 correlated with negative PC2 values (Fig. [98]1E).
This is in line with the typically less aggressive behavior of CLL with
wild-type NOTCH1 [[99]20, [100]21].
We also found that the viability of these primary CLL cells was
positively correlated with PC1 and PC2, both of which have decreased
apoptotic priming at their positive values, following 48 h of culturing
in media (Fig. [101]1F). This validates our BH3-profiling data as a
robust measurement of the propensity of CLL cells to undergo apoptosis,
and also suggests that samples that are sensitive to BH3
peptide-induced CytC release (negative PC1 or 2 values) are likely to
undergo cell death, as they are more readily primed for apoptosis and
potentially more susceptible to therapeutic intervention.
BCL-2 dependence is positively correlated with known prognostic markers in
CLL
Given that the PCA demonstrated correlations with mutational status of
CLL, we next sought to evaluate the correlation between dependence on
specific anti-apoptotic proteins and established genomic markers
(Fig. [102]2A, S4). The most significant association was observed
between the prognostically favorable mutated IGHV [[103]17] and higher
BCL-2 dependence, followed by trisomy 12 and higher BCL-2 dependence
(Fig. [104]2A, B, S4). Lower BCL-2 dependence also correlated with CLL
cells harboring the poor prognostic marker NOTCH1 mutation
(Fig. [105]2B) [[106]20–[107]22]. Notably, pre-treatment status was not
a confounding factor when included as a co-variate in our linear models
for association, as the association tests results were largely similar
to when only previously untreated patients were included (Fig. S5). We
next performed multivariate regression with LASSO penalty to select
genomic and demographic features that can explain the heterogeneity of
BH3 profiles. Among the selected features, poor prognostic factors such
as del(17p) and NOTCH1 mutation were correlated with lower BCL-2
dependence, while more favorable prognostic factors, such as del(13q)
and trisomy 12 [[108]23, [109]24], were correlated with higher BCL-2
dependence (Fig. [110]2C). Conversely, only MCL-1, with lower
dependence, was correlated with del(13q) (Fig. [111]2C). These findings
suggest that anti-apoptotic dependence, particularly BCL-2 dependence,
could be favorable predictive biomarkers, based on correlation with
various prognostic genetic biomarkers. Interestingly, the predictive
potential of BCL-2 dependence was further supported by the correlations
between relevant apoptotic gene expressions and established genomic
markers (Fig. [112]2D). For example, BCL2 expression was higher in
mutated IGHV or non-del(17p) CLL samples. Relevantly, these samples
also express higher pro-apoptotic genes such as PMAIP (NOXA), BAX or
BCL2L11 (BIM), indicating that these BCL-2 dependent CLL cells are
again more readily primed for apoptosis due to the higher expression of
pro-apoptotic genes. PMAIP expression was also higher in del(13q)
cells, which aligns with our earlier observation of lower MCL-1
dependence in del(13q) cells, as NOXA is known to degrade MCL-1
[[113]25, [114]26] (Fig. [115]2D). This serves as a control to indicate
that not all dependence, but BCL-2 dependence specifically, is
associated with favorable genomic markers.
Fig. 2.
[116]Fig. 2
[117]Open in a new tab
Specific genetic determinants through BH3-profiling of CLL cells.
A Summary of significant associations between genetic variations and
BH3 profiles (Student’s t-test). Dashed line is 10% FDR. B Boxplot of
significant associations between BH3 profile and genetic variations.
Only associations passed 10% FDR control are shown. (BAD, ABT199,
PUMA = BCL-2 dependence, FS1 = BFL-1 dependence). C Genetic and
demographic features selected by regularized multi-variate linear
regression (LASSO) for explaining BH3 profile pattern. Horizontal bars
on the left indicate the average feature coefficients in the regression
models and scatter plot at the bottom indicates z-score of CytC release
following peptide treatment. D Boxplot of significant associations
between BCL2 family gene expressions and genetic variations. Only
associations above the 10% FDR control are shown. (BCL2 – BCL-2; MCL1 –
MCL-1; BAX – BAX; BCL2L11 – BIM; PMAIP – NOXA)
To further evaluate the predictive potential of BCL-2 dependence, we
analyzed the correlation between BH3 profiles and prognostic outcomes.
We found that only ABT199 and BAD (corresponding to BCL-2 dependence)
were significantly associated with time to treatment (TTT) in our
univariate analysis (P value = 0.012 for BAD and 0.030 for
ABT199) (Fig. S6). Overall survival (OS) data were not mature enough to
be evaluated, as only 8 out of 73 patients were deceased in the cohort.
High BCL-2 dependence is associated with gene expression patterns and
functional pathways that reflect favorable treatment response
Apart from our robust correlations between BCL-2 dependence and
favorable prognostic markers, our PCA previously suggested that cells
that are sensitive to BH3 peptide-induced CytC release were more
readily primed for cell death, hence suggesting more responsiveness to
treatment intervention (Fig. [118]1F). To further investigate this
finding, we used available clinical data from 6 patients with CLL
previously treated with a purine-analog based chemoimmunotherapy
combination of pentostatin, cyclophosphamide, rituximab (PCR) to
examine whether priming or specific anti-apoptotic protein dependence
could predict for response to treatment. As one measure of clinical
activity, we studied the rate of change of absolute lymphocyte counts
(ALC) during the course of this therapy. We observed that BCL-2
dependence (R = 0.89, P = 0.0161), and to a lesser extent BFL-1
dependence (R = 0.86, P = 0.0274), were positively correlated with the
lymphocyte count drop rate after treatment (Fig. [119]3A). This finding
provided preliminary support that anti-apoptotic protein dependence,
particularly on BCL-2, could predict for a favorable treatment
response.
Fig. 3.
[120]Fig. 3
[121]Open in a new tab
Specific gene expression, signaling-pathway patterns and in vivo drug
response are correlated with BCL-2 dependence. A Associations between
BH3 profile and in vivo drug response estimated from retrospective
clinical data (lymphocyte drop rate) (Pearson’s correlation test). The
lymphocyte drop rate was calculated by dividing the change in
lymphocyte count (in millions) between the start and end of therapy by
the duration of the therapy (in days) for each patient. B Example of
significant correlations between BCL-2 dependence (reflected by ABT199,
BAD or PUMA-induced CytC release) and mRNA expressions of PMAIP1
(NOXA), BCL2L11 (BIM), COX5A, LAG3, measured by RNA-seq. C Cancer
hallmark pathways enrichment analysis for mRNAs correlated with BCL-2
dependence. Enrichment results passed 10% FDR control are shown. D A
heatmap showing the mRNA expression of the genes (Blue to red bar
indicates expression levels), both associated with BCL-2 dependence
(Green bar indicates increasing BCL-2 dependence) and belong to the
TNFα/NFκB-signaling pathway
To further understand the functional aspects of anti-apoptotic protein
dependence as a predictive biomarker for treatment response, we
evaluated the association between gene expression levels and dependence
on various anti-apoptotic BCL-2 family members. We found that only
BCL-2 dependence (ABT199 or BAD) was associated with differential RNA
expression (Fig. S7-S9). Importantly, we observed positive correlations
between BCL-2 dependence and expression of BCL2L11 (BIM) and PMAIP1
(NOXA)(Fig. [122]3B, S8-S9), which were among the strongest
correlations in our analyses. Notably, BIM and NOXA proteins are a
pro-apoptotic sensitizer and activator, respectively, that promote
apoptotic-cell death. Other pro-cell death genes were also positively
correlated with BCL-2 dependence, such as the inter-strand
crosslink-repair protein, FANCI and androgen transmembrane protein,
PMEPA1 [[123]27–[124]29]. In contrast, BCL-2 dependence correlated
inversely with expression of COX5A, LAG3, MYLK (Fig. [125]3B, S8-S9),
genes that are related to CLL and cancer cell survival or progression
[[126]30–[127]32]. These observations suggest that CLL cells with high
BCL-2 dependence undergo apoptosis more readily and therefore may be
more susceptible and responsive to drug treatment.
Pathway enrichment analysis further suggest that BCL-2 dependence is
positively correlated with TNFα/NFκB-signaling, while inversely
correlated with oxidative phosphorylation (OXPHOS) (Fig. [128]3C, D).
The positive correlation between BCL-2 dependence and the TNFα/NF
[MATH: κ :MATH]
B-regulated mitochondrial anti-oxidant superoxide dismutase-2 (SOD2),
indicates that with higher SOD2 expression, more intracellular
superoxide (O[2]^.−) is removed to induce a lower redox milieu in
BCL-2-dependent cells (Fig. [129]3D). This supports a more
drug-sensitive phenotype, as increased reactive oxygen species (ROS),
O[2]^.− or redox milieu have been implicated in drug resistance,
improved cancer cell survival and proliferation [[130]31,
[131]33–[132]36]. Other notable gene expression under TNFα/NF
[MATH: κ :MATH]
B-signaling is ATPase plasma membrane Ca^2+ transporter-1, ATP2B1
(Fig. [133]3D), previously reported to promote apoptosis [[134]37,
[135]38].
The inverse correlation between BCL-2 dependence and OXPHOS from our
pathway enrichment (Fig. [136]3C) is also consistent with the inverse
correlation between BCL-2 dependence and the mitochondrial complex-IV
subunit gene, COX5A (Fig. [137]3B), which positively affects
mitochondrial respiration (OXPHOS) and redox metabolism [[138]31,
[139]33]. Given that high OXPHOS and mitochondrial O[2]^.− are
implicated in drug resistance, for example in CLL cells resistant to
ABT199 (VEN) [[140]36], our finding of high BCL-2 dependence
correlating with low COX5A, OXPHOS and high SOD2 provides some initial
mechanistic insight into the reason why BCL-2 dependence is a
predictive biomarker of drug sensitivity. Thus, these data suggest that
high BCL-2 dependence could be a functional biomarker that predicts for
a favorable treatment response, due to its correlation with
drug-sensitive phenotypes.
High BCL-2 dependence is an underlying mechanism of increased drug
sensitivity
To better understand the mechanistic underpinnings of our hypothesis
that BCL-2 dependence predicts favorable treatment response, we devised
an experimental system utilizing two diffuse large B-cell lymphoma
(DLBCL) cell lines – one with high dependence on BCL-2 (OCI-Ly1) and
another with lower dependence on BCL-2 (OCI-Ly3), demonstrated by BAD-
and ABT199-induced CytC release and its delta CytC release (values of
CytC release derived from CytC release values of OCI-Ly3 minus that of
OCI-Ly1 cells) (Fig. [141]4A, B). Of note, cell lines more typically
utilized in other CLL studies such as OSU-CLL, HG3, and MEC-1, were not
used here, as these cell lines generally have markedly low dependence
on BCL-2 as compared to primary CLL cells, and are therefore relatively
resistant to BCL-2 inhibition as compared to primary CLL cells (Fig.
S10) [[142]39, [143]40]. Hence, the two DLBCL lines with drastically
different BCL-2 dependences were selected as experimental tools to
delineate the different responses toward drug treatment(s), since one
line is highly BCL-2 dependent (like primary CLL cells) and the other
is distinctly not BCL-2 dependent, but is otherwise similar to its
BCL-2 dependent cell line counterpart.
Fig. 4.
[144]Fig. 4
[145]Open in a new tab
Correlations between BCL-2 dependence and multi-omics analyses were
experimentally re-capitulated in cell lines. A Baseline BH3-profiling
was performed for two DLBCL cell lines OCI-Ly1 and OCI-Ly3. High CytC
release by BH3 peptides or mimetic drug indicates inclined dependence
towards its respective anti-apoptotic protein(s) for survival (n = 3).
Specific dependences as reflected on the right. B Delta CytC release
between OCI-Ly3 and OCI-Ly1 (CytC release % value of OCI-Ly3 minus CytC
release % value of OCI-Ly1) was calculated to reflect absolute changes
in dependence to specific anti-apoptotic protein(s) for survival.
Net + % indicates relative increased or better dependence and net—%
suggests relative decreased dependence (n = 3). C Western blot analysis
showing protein expression levels of BIM-xL, L, S, phosphor-IKK, total
IKK, BAX, BAK and respective β-Actin(s) of OCI-Ly1 and OCI-Ly3.
D Mitochondrial respiration or OXPHOS measurement between OCI-Ly1 and
OCI-Ly3. Asterisks indicate addition of oligomycin (*), FCCP (**),
antimycin-A/rotenone (***) to indicate basal and maximal respirations.
Representative graph is shown here (n = 3). E Mitosox staining was
measured and quantified between OCI-Ly1 and OCI-Ly3 (n = 3). Unpaired
t-test was used. F, G Cell viability by CellTiter-Glo® was measured
between OCI-Ly1 and OCI-Ly3 following 48-h treatment with increasing
doses of ABT199 (VEN) (n = 4), ibrutinib (n = 3), etoposide (n = 4) or
doxorubicin (n = 4)
We found that OCI-Ly1 (higher BCL-2 dependence) cells, relative to
OCI-Ly3 (lower BCL-2 dependence) cells, have higher BIM protein
expression and TNFα/NF
[MATH: κ :MATH]
B-signaling by Western blot analysis (Fig. [146]4C), re-capitulating
our transcriptomic analyses of primary CLL cells (Fig. [147]3B-D).
Consistent with our transcriptomic data showing lower COX5A and higher
SOD2 expression in highly BCL-2-dependent primary CLL cells
(Fig. [148]3B, D), we observed lower mitochondrial respiration/OXPHOS
and mitochondrial O[2]^.− levels in OCI-Ly1 compared to OCI-Ly3
(Fig. [149]4D, E). Importantly, OCI-Ly1 cells possessed higher
pro-apoptotic BAX and BAK protein expression (Fig. [150]4C) and were
more sensitive to various targeted and chemotherapeutic drug
treatments, including ibrutinib, ABT199 (VEN), doxorubicin and
etoposide relative to OCI-Ly3 cells (Fig. [151]4F, G). These findings
in DLBCL cell lines reflect the phenotypes and potential treatment
response of primary CLL cells with different degrees of BCL-2
dependence.
To further examine if these associations with BCL-2 dependence were
causal, we proceeded to investigate the therapeutically and immediately
actionable targets or functions from transcriptomic and pathway
enrichment analyses, namely OXPHOS and mitochondrial O[2]^.−
production, which are associated with BCL-2 dependence. We treated the
relatively drug-resistant and lower BCL-2-dependent OCI-Ly3 cells with
an inhibitor of NAPDH oxidase-dependent O[2]^.− production,
diphenyleneiodonium (DPI), or a mitochondrial complex-III/OXPHOS
inhibitor, antimycin-A (AA), and observed that this led to an increase
in BCL-2 dependence (Fig. [152]5A), and a clear switch of net negative
to positive BCL-2 dependence when comparing DPI or AA with DMSO control
(Fig. [153]5B, delta values derived from drug-treated minus control
groups). We recently reported that BCL-2 phosphorylation negatively
regulates the dependence of BCL-2 in DLBCL and CLL cells [[154]13]. Our
data further confirm that AA and DPI could reduce BCL-2 phosphorylation
in OCI-Ly3 cells (Fig. [155]5C), thus corroborating the increased
dependence on BCL-2 for survival. DPI is also used here as a positive
control for the reduction in BCL-2 phosphorylation [[156]41]. The
decrease in BCL-2 phosphorylation by AA/DPI alone corresponds to a
slight drop in cell survival (Fig. [157]5D). Importantly, increase in
BCL-2 dependence was further accompanied by a significant increase in
sensitization to chemo- and targeted drugs doxorubicin, etoposide,
ABT199 (VEN) and ibrutinib (Fig. [158]5D). These findings thus validate
the associations between BCL-2 dependence and multi-omics analyses in
primary CLL cells and confirm that high BCL-2 dependence is one
mechanism underlying improved treatment response.
Fig. 5.
[159]Fig. 5
[160]Open in a new tab
BCL-2 dependence is a causal function of drug sensitivity in cancer
cells in vitro. A BH3-profiling performed following treatment with DPI
(0.1 μM) or AA (0.05 μM) for 15 h for OCI-Ly3 cells (n = 3). B Delta
CytC release between DPI/AA- and DMSO-treated OCI-Ly3 (CytC release %
value of DPI/AA minus CytC release % value of DMSO) was calculated to
reflect absolute changes in dependence to specific anti-apoptotic
protein(s) for survival (n = 3). C Western blot analysis showing
S70pBCL-2, BCL-2 and β-Actin levels of OCI-Ly3. D Cell viability by
CellTiter-Glo® was measured for OCI-Ly3 following 2-h pre-treatment
with DPI (0.1 μM) or AA (0.05 μM) and subsequent 48-h co-treatment with
either doxorubicin (0.1 μM, n = 4), etoposide (0.1 μM, n = 4), ABT199
(VEN) (0.1 μM, n = 4) or ibrutinib (1 μM, n = 3). Tukey’s multiple
comparisons test was used
To corroborate these observations from cell lines, we profiled 8
additional primary CLL samples and clustered them based on their
degrees of BCL-2 dependence (patients 1–8). We observed that patients
5–8 had relatively higher BCL-2 dependence as compared to patients 1–4
(Fig. [161]6A). We further observed that patients 5–8 also had higher
pro-apoptotic BIM and BAX protein expressions (Fig. [162]6B). Notably,
strong and significant correlations between BCL-2 and BIM (r = 0.9011,
P = 0.0022) or BCL-2 and BAX (r = 0.8166, P = 0.0134), but not MCL-1
and BIM (r = 0.3919, P = 0.3369) or MCL-1 and BAX (r = 0.1708,
P = 0.686), were observed (Fig. [163]6C, D). Importantly, BCL-2
dependence, but not MCL-1 dependence (Fig. S11), was highly associated
with BCL-2 (r = 0.8166, P = 0.0138), BAX (r = 0.8138, P = 0.014) and
BIM (r = 0.882, P = 0.0038) protein expressions (Fig. [164]6E). These
data again corroborate our correlative data on BCL-2 dependence and
RNA-Seq (Fig. [165]3B), as well as our recent study on defining
anti-apoptotic protein dependences by specific exogenous BCL-2 family
protein expression [[166]13], and support the hypothesis that
BCL-2-dependent cells are more primed for apoptosis due to the higher
levels of pro-apoptotic proteins.
Fig. 6.
[167]Fig. 6
[168]Open in a new tab
High BCL-2 dependence predicts favorable drug response in primary CLL
cells ex vivo. A Baseline BH3-profiling was performed on primary CLL
cells. Cells were incubated in specific BH3 peptides or mimetic for 1 h
prior to determining CytC loss. Patients arbitrarily labeled as patient
1–8. B Western blot analysis showing MCL-1, BCL-2, BIM xL, L, S, BAX
and β-Actin levels of the same 8 CLL patient samples in Fig. 6A. C,
D Pearson correlation coefficient analyses between BCL-2 family protein
expressions, normalized to β-Actin levels from Fig. 6B. E Pearson
correlation coefficient analyses of BCL-2 dependence and normalized
BCL-2, BAX or BIM protein expression levels. F Replot of baseline
BH3-profiling of primary CLL cells from patient 1 and 7 with different
degrees of MCL-1 and BCL-2 dependences. G Cell death levels by fold
change (F.C.) of annexin V staining of the same patient 1 and 7
following 24-h ex vivo fludarabine or ibrutinib treatment (with stroma
NKtert co-culture) (Viability was measured in duplicates of the same
patient, n = 2). H BH3-profiling performed following ex vivo treatment
with DPI or AA for 6 h in 8 CLL patient primary cells to evaluate BCL-2
dependence level. I Cell death levels by fold change of annexin V
staining of CLL patient primary cells following 2-h pre-treatment with
AA or DPI followed by 24-h ibrutinib (n = 8) (with stroma NKtert
co-culture). Sidak’s multiple comparisons test was used. J Western blot
analysis showing S70pBCL-2, BCL-2 and β-Actin levels of 8 CLL patient
primary cells following ex vivo treatment with DPI or AA for 6 h. A
representative patient shown. Densitometry analysis is in supplemental
figure S12C. K BH3-profiling performed following ex vivo treatment of
0.1 μM IACS-010759 for 6 h in 8 CLL patient primary cells to evaluate
BCL-2 dependence level. L Cell death levels by fold change of annexin V
staining of CLL patient primary cells following 2-h pre-treatment with
0.1 μM IACS-010759 followed by 24-h 2.5 nM ABT199 (VEN) (n = 8) or 1 μM
ibrutinib (n = 7) (with stroma NKtert co-culture). Sidak’s multiple
comparisons test was used
To further investigate the correlation between BCL-2 dependence and
treatment response, we analyzed patients 1 and 7, who both had samples
with sufficient cell numbers but also had opposite anti-apoptotic
dependences (Fig. [169]6F) We treated these cells with
a chemotherapeutic agent, fludarabine, or targeted agent, ibrutinib ex
vivo and measured cell viability (Fig. [170]6G). We observed again that
primary CLL cells from patient 7 with higher BCL-2 dependence were more
susceptible to apoptotic cell death when treated with fludarabine or
ibrutinib, as compared to those of patient 1 with lower BCL-2
dependence (Fig. [171]6F, G). In contrast, the higher MCL-1 dependence
in patient 1's cells did not respond as well, hence indicating that
drug sensitivity is not arbitrarily governed by any high anti-apoptotic
protein dependence (Fig. [172]6F, G).
Finally, we treated cells from 8 additional patients with CLL of
varying BCL-2 dependences with AA or DPI ex vivo, and observed that
these drugs could similarly increase BCL-2 dependence (Fig. [173]6H,
S12A), which further translated to enhanced apoptotic cell death when
combined with ibrutinib or fludarabine (Fig. [174]6I, S12B). We
demonstrated that BCL-2 dependence and cell death were again enhanced
by the decrease in BCL-2 phosphorylation in primary CLL cells
(Fig. [175]6J, S12C). To ensure reproducibility and clinical relevance,
we studied an OXPHOS inhibitor, IACS-010759, that has undergone
clinical trials [[176]42]. Indeed, IACS-010759 was capable of
increasing BCL-2 dependence in CLL patient samples as well as
sensitivity to both targeted agents, ABT199 (VEN) and ibrutinib
(Fig. [177]6K, L, S12D). Collectively, our in vitro and ex vivo data
demonstrate consistently that higher BCL-2 dependence underlies better
response to various treatment modalities.
Functional BCL-2 dependence predicts clinical response of CLL patients to a
targeted chemoimmunotherapy-based regimen
Having mapped the landscape of functional BCL-2 family dependence
across a range of CLL genetic backgrounds with transcriptomic and
proteomic analyses, and having performed experimental validation to
unravel the causal relationship between BCL-2 dependence, pro-cell
death gene expressions, drug sensitivity and response, we next applied
this technique to a validation cohort of CLL samples from 54 patients
treated uniformly on a prospective, phase-2 trial of first-line
ibrutinib plus FCR (iFCR) [[178]14]. Our aim was to validate that with
higher BCL-2 dependence, patients will be more responsive to treatment.
We generated two datasets by BH3-profiling these samples: (i) baseline
CytC release data of pretreatment patient samples, and (ii) delta
values from CytC release data of paired patient samples after 1 week of
ibrutinib monotherapy minus CytC release data of pretreatment patient
samples (Fig. [179]7A). The latter dataset was assessed to inform how
ibrutinib alters BH3-profiling results. To characterize the
anti-apoptotic dependence patterns of these two datasets, we initially
performed an unsupervised hierarchical clustering analysis (Fig. S13A).
Similar to our initial discovery cohort, we observed that CLL patients
whose CLL cells had high BCL-2 dependence at baseline as well as CLL
patients with a further net increase in BCL-2 dependence following
ibrutinib treatment (positive delta CytC values) were both associated
with the more favorable mutated IGHV status (Fig. [180]7B).
Importantly, results from the 1 week of ibrutinib dataset confirmed
that as long as BCL-2 dependence is high, CLL cells will still be
associated with favorable genetic markers such as mutated IGHV.
Fig. 7.
[181]Fig. 7
[182]Open in a new tab
BCL-2 dependence predicts for complete remission and undetectable
minimal residue disease in validation cohort. A Paired blood samples
from patients with CLL were drawn pre-treatment and after 1 week of
ibrutinib therapy in a frontline iFCR phase-2 clinical trial. PBMCs
from these patients were isolated and used directly for in vivo
BH3-profiling. Diagram was created in BioRender. Chamberlain, S. (2025)
[183]https://BioRender.com/h74g455. B Associations between the levels
of CytC release by ABT199 (BCL-2 dependence) and IGHV mutational status
in baseline patients as well as association between the delta levels of
CytC release by ABT199 (BCL-2 dependence) and IGHV mutational status in
in vivo ibrutinib-treated patients. Positive delta values from in vivo
ibrutinib group indicates further net increase in BCL-2 dependence
following ibrutinib treatment in comparison to baseline screening and
vice versa for negative delta values. C P value heatmap plot
summarizing the P values of association between BH3 profile and various
clinical outcomes. Positive P values (red) indicating higher CytC
release in “neg” group for MRD or “PR” group for response. MRD -
minimal residue disease, BM - bone marrow, PB - Peripheral Blood.
D Boxplots showing the examples of associations between baseline or
delta BCL-2 dependence (represented by ABT199) and clinical outcomes.
The P values were calculated by Student’s t-test. E Boxplots showing
the examples of IGHV-independent associations between baseline or delta
BCL-2 dependence and clinical outcomes. The P values were calculated by
two-way ANOVA tests including IGHV mutational status as an additional
covariate. F Boxplot of significant associations between protein
expressions of paired CLL patient samples before and after in vivo
ibrutinib therapy additional cohort (n = 5). Only associations above
the 10% FDR control are shown
Next, we correlated anti-apoptotic dependence data generated through
BH3-profiling with the clinical responses (complete response, CR vs.
partial response, PR) and minimal residual disease (MRD) status halfway
through combination therapy (post-cycle 3), at the end of combination
therapy (post-cycle 6), and at time of best response on trial
(including patients who went on to receive ibrutinib monotherapy
maintenance). We hypothesized that BCL-2 dependence would predict depth
of clinical response in the iFCR clinical trial cohort [[184]14]. We
observed that there were significant correlations between the different
peptides/BH3 mimetic at various endpoints. Specifically, higher BCL-2
dependence in CLL cells from patients at baseline was indeed associated
with increased likelihood of achieving undetectable MRD (uMRD) and CR
at the halfway mark on treatment, at the end of combination therapy and
at the time of best response (Fig. [185]7C, D). Additionally, CLL cells
with further net increase in BCL-2 dependence (positive delta CytC
values) after 1 week of ibrutinib treatment were similarly associated
with increased likelihood of achieving uMRD and CR at these endpoints
(Fig. [186]7C, D, S13B). These findings were consistent with our
initial observations of the 6 PCR-treated patients (Fig. [187]3A) as
well as our in vitro and ex vivo studies (Figs. [188]4– [189]6).
Interestingly, both BFL-1 and BCL-xL dependence were also associated
with response and uMRD status in the in vivo ibrutinib-treated group,
though these associations were only apparent at earlier timepoints
(Fig. [190]7C). In our uni- and multivariate analyses, the association
between BCL-2 dependence and uMRD was not confounded by IGHV mutational
status (Fig. [191]7E), and IGHV mutational status alone did not predict
uMRD at any specified timepoint (Fig. S14). This suggests that BCL-2
dependence itself is a primary factor associated with likelihood of
achieving uMRD, supporting our overall hypothesis that BCL-2 dependence
is a potential independent biomarker of favorable response to various
therapies in CLL. In addition, proteomics data from an additional,
independent in vivo ibrutinib-treated patient cohort demonstrated that
ibrutinib could increase BCL-2, BIM, SOD2, while decrease COX5A/B
(Fig. [192]7F), proteins corresponding to our earlier observations that
support BCL-2 dependence in association with better apoptotic priming
and treatment response (Fig. [193]3B, [194]4C, [195]6B). This further
supports our finding that ibrutinib increases BCL-2 dependence in our
ibrutinib-treated cohort (Fig. [196]7C), and corroborates our previous
study demonstrating that ibrutinib increases BCL-2 dependence
[[197]12]. Finally, to investigate the reliability of the BH3-profiling
test in predicting treatment response and thus the validity of BCL-2
dependence in predicting favorable treatment response, we calculated
area under the receiver operating characteristic (AUROC) curves for
each of the BH3 features when predicting each clinical outcome. We
displayed the top 5 features from baseline or in vivo ibrutinib
treatment (Tables [198]1 and [199]2, complete list in supplementary
material 1). We observed that these top features from both baseline and
in vivo ibrutinib sets with good to excellent prediction powers were
all features of BCL-2 dependence; for example, baseline ABT199 at 1 μM
(AUROC = 0.802 ± 0.082) and in vivo PUMA peptide at 1 μM
(AUROC = 0.827 ± 0.127). In contrast, features of BCL-xL or BFL-1
dependence were displayed at lower rankings (Supplemental Table 1). The
AUROC analysis therefore provides validation of BH3-profiling and
confirms our observations that BCL-2 dependence is a potential
predictive biomarker of favorable response for different treatments in
CLL.
Table 1.
AUROC of baseline BCL-2 dependence for the prediction of treatment
response and outcome
FEATURE AUROC (VALUE) AUROC
(SD) TREATMENT TIMEPOINT RESPONSE (AREA)
ABT199 (1 μM) 0.802 0.082 Post Cycle 3 MRD PB
ABT199 (1 μM) 0.783 0.117 Post FCR MRD PB
ABT199 (0.1 μM) 0.773 0.118 Post FCR MRD PB
BAD (0.1 μM) 0.760 0.112 Post FCR MRD PB
ABT199 (0.5 μM) 0.752 0.133 Post FCR MRD PB
[200]Open in a new tab
Post Cycle 3: Halfway through combination therapy
Post FCR: End of combination therapy
MRD PB: Minimal Residue Disease Peripheral Blood
Table 2.
AUROC of BCL-2 dependence following ibrutinib treatment for the
prediction of treatment response and outcome
IN VIVO IBRUTINIB
FEATURE AUROC (VALUE) AUROC
(SD) TREATMENT TIMEPOINT RESPONSE (AREA)
PUMA (1 μM) 0.827 0.127 Post FCR MRD PB
ABT199 (0.1 μM) 0.779 0.171 Post FCR MRD PB
BAD (1 μM) 0.778 0.125 Best Response MRD PB
ABT199 (0.1 μM) 0.774 0.130 Best Response MRD BM
ABT199 (1 μM) 0.758 0.149 Post FCR MRD PB
[201]Open in a new tab
Post FCR: End of combination therapy
MRD PB: Minimal Residue Disease Peripheral Blood
MRD BM: Minimal Residue Disease Bone Marrow
Discussion
We used BH3-profiling to unlock the key molecular features defining
anti-apoptotic protein dependence and response to treatment in CLL. In
our discovery cohort of 73 CLL patients, we found that higher BCL-2
dependence was associated with favorable genetic prognostic factors and
lesser reliance on OXPHOS and mitochondrial O[2]^.−. In contrast, lower
BCL-2 dependence was associated with poor prognostic factors, high
OXPHOS and mitochondrial O[2]^.− reliance. BCL-2 dependence was
correlated with transcriptomic signatures that indicate a more
apoptotic-ready and drug-sensitive milieu. We further confirmed that
these correlations were underlying regulatory factors for BCL-2
dependence and subsequent drug response in cell lines and ex vivo
patient samples. In our uniformly-treated iFCR clinical trial, we
ascertained that primary CLL cells from patients at baseline or after
1 week of ibrutinib treatment with higher BCL-2 dependence in vivo
independently predicted depth of clinical response, while IGHV
mutational status alone was unable to do so. The latter analysis is
consistent with our recent publication that clinical response in
patients with CLL treated with iFCR was not affected by IGHV mutational
status [[202]43]. Overall, our study is the first to comprehensively
define BCL-2 dependence as a potential functional and predictive
treatment response biomarker in CLL, integrating multi-omics data and
BH3-profiling to validate its clinical utility.
Our analysis not only demonstrated that BCL-2 dependence and associated
expression of apoptotic genes correlated positively with favorable
prognostic biomarkers, but also identified associations of BCL-2
dependence with drug response and a drug-sensitive transcriptome, which
collectively provided us an angle to investigate the possibility of
BCL-2 dependence in predicting treatment response in CLL patients. For
example, higher BCL-2 dependence was associated with increased
expression of the pro-apoptotic BH3-only protein NOXA. NOXA is known to
prevent the anti-apoptotic MCL-1 from binding to the pro-apoptotic
activator BIM and effector BAK [[203]25, [204]44–[205]46]. It has also
been suggested that NOXA may increase proteasomal degradation of MCL-1
[[206]25, [207]26], which leads to increased sensitivity to BCL-2
inhibition [[208]13, [209]36]. Hence, increased NOXA expression may
reduce MCL-1 protein, increasing CLL cell dependence on BCL-2 for
survival. Indeed, NOXA genetic amplification has recently been
suggested to increase sensitivity to BCL-2 inhibitors in DLBCL
[[210]47]. Importantly, we found that BCL-2 dependence, together with
BCL-2 protein expression, were positively associated with BAX/BAK and
BIM expression. Higher BAX, BAK and BIM expression suggest that the
impact of the canonical role of BCL-2 in sequestering and inhibiting
BIM/BAX/BAK-induced apoptosis may increase, thus translating to an
enhanced dependence on BCL-2 for survival. When these pro-apoptotic
protein levels are high, in association with high BCL-2 and its
dependence for survival, this indicates that BCL-2 is loaded with
pro-apoptotic proteins, and when triggered by drugs, may release a
floodgate of active pro-apoptotic effectors to induce apoptotic cell
death. This provides a mechanistic rationale for our finding that BCL-2
dependence, in positive association with BAX/BAK and BIM expression,
predicts better response to treatment. Our findings thus collectively
begin to elucidate the mechanisms underlying high BCL-2 dependence in
CLL, and how BCL-2 dependence may predict for treatment response.
BCL-2 dependence also positively correlated with TNFα/NFκB-signaling.
While TNFα/NFκB-signaling is frequently associated with cell survival,
evolving evidence has implicated its involvement in cell death
regulation [[211]37, [212]38, [213]48–[214]51]. For example,
TNFα-induced apoptosis may first be inhibited by the activation of
Rel/NFκB, but over time increases spontaneous apoptosis in cancer
cells. This was shown to be due to the upregulation of SOD2, which
serves as an anti-oxidant protein to convert O[2]^.− to hydrogen
peroxide (H[2]O[2]) [[215]49]. Our findings demonstrating higher BCL-2
dependence in associated with higher SOD2 expression and lower
mitochondrial O[2]^.− levels in vitro support this prior observation.
We hypothesize that tumor cells may survive due to SOD2 upregulation,
which rapidly detoxifies the initial apoptogenic burst of
drug/TNFα-induced O[2]^.− to H[2]O[2], but they eventually succumb to
the progressive build-up of H[2]O[2] that leads to apoptosis [[216]49].
Interestingly, TNFα and NFκB have both been shown to positively
regulate BCL-2 transcription, and we therefore anticipate that an
increase in BCL-2 dependence is due in part to an increase in BCL-2
transcription by TNFα/NFκB signaling [[217]52, [218]53].
The inverse correlation between BCL-2 dependence and OXPHOS from our
enrichment pathway analysis is also supported by our in vitro data. As
high OXPHOS and the subsequent increase in mitochondrial O[2]^.−
production have been implicated in resistance to various therapeutic
drugs [[219]31, [220]33, [221]34, [222]36, [223]41, [224]54], the lower
OXPHOS and mitochondrial O[2]^.− levels, and consequentially higher
BCL-2 dependence, suggest a more drug-sensitive milieu and favorable
treatment response, as observed in our experimental data in cell line
and primary CLL samples. We previously demonstrated that COX5A/O[2]^.−
could activate the anti-apoptotic activity of BCL-2 via its
phosphorylation [[225]31, [226]35] and that BCL-2 phosphorylation could
suppress BCL-2 dependence [[227]13]. Here, we confirmed that inhibiting
OXPHOS or preventing O[2]^.− production could reduce BCL-2
phosphorylation and reciprocally increase BCL-2 dependence and drug
sensitivity. Interestingly, TNFα/NFκB is known to positively regulate
both SOD2 and BCL2 transcriptions, and previous work demonstrated that
SOD inhibition increases O[2]^.− levels, thereby increasing BCL-2
phosphorylation and reducing drug sensitivity [[228]41]. These previous
findings corroborate our work here, and provide a potential link
between TNFα/NFκB signaling and BCL-2 phosphorylation via SOD2. We
further hypothesize that the increase in TNFα/NFκB signaling would
increase expression of both SOD2 and BCL-2, potentially allowing more
SOD2 to reduce O[2]^.− levels and BCL-2 phosphorylation, ultimately
making cells more dependent on BCL-2 and sensitive to drug treatments.
Collectively, our findings support the observation that a favorable
treatment response to therapy is more likely when CLL cells have higher
BCL-2 dependence, and our data provide initial insights into the
mechanisms underlying BCL-2 dependence.
Our study further suggests that BCL-2 dependence should be explored as
a predictive biomarker to identify CLL patients who are more likely to
respond favorably not only to conventional chemotherapy but also to
targeted therapy. We observed that CLL cells with higher BCL-2
dependence, treated with targeted agents such as ABT199 (VEN),
ibrutinib or conventional chemotherapy such as fludarabine,
doxorubicin, etoposide, had higher cell death, and that CLL patients
treated with iFCR were likely to have improved clinical response. Our
observations here support exploring alternative therapeutic approaches
for CLL patients with lower levels of BCL-2 dependence. Such patients
may benefit from a combination treatment approach incorporating drugs
targeting one or more of the therapeutically actionable nodes, which
may increase BCL-2 dependence. One such example is the use of protein
phosphatase 2A activator drugs (PADs) [[229]13, [230]55], that we
recently demonstrated increase BCL-2 dependence through the activation
of protein phosphatase 2A and reduction of BCL-2 phosphorylation
[[231]13]. Given that our transcriptomics analysis also provided
multiple pharmacologically-actionable targets that are linked to lower
BCL-2 dependence (Fig. [232]3B, C, S8-S9), this suggests several new
potential combination partners that could increase BCL-2 dependence and
thereby sensitize CLL cells to various targeted and/or chemotherapy
regimens. For example, targeted treatment combinations with low doses
of OXPHOS inhibitors (including several currently under clinical
evaluation) [[233]42]) or ROS scavengers (i.e. SOD2 mimetic,
GC4419) [[234]56] with ibrutinib and/or ABT199 (VEN) could be explored.
Thus, in addition to BCL-2 dependence serving as a potential predictive
biomarker of treatment response, understanding the mechanisms
underlying its modulation may help to potentially identify new
therapeutic combinations.
It is worth noting the possibility that some CLL cells may depend more
on MCL-1 than on BCL-2. We observed in a previous study that CLL and
DLBCL cells with lower BCL-2 dependence, display elevated MCL-1
dependence, which translates to higher sensitivity to MCL-1 inhibitor,
[235]S63845 [[236]13]. Nonetheless, as toxicity issue has to date
impeded the clinical development of MCL-1 inhibitors, our data here
provide a potential alternative to increase BCL-2 dependence in CLL
patients with initially lower BCL-2 dependence. Another important
factor worth mentioning is the ability of ibrutinib to induce tumor
cell release from lymph nodes [[237]57], raising the question of
whether the in vivo ibrutinib treatment in our iFCR cohort would
influence the BH3-profiling results. Similar to our baseline profiling
showing the association between BCL-2 dependence and better treatment
response, the 7-day treatment with ibrutinib showed very similar
results, in which our profiling at this timepoint showed that higher
BCL-2 dependence is associated with better treatment response. As such,
the ability of ibrutinib to release tumor cells from the lymph nodes
does not seem to affect our BH3-profiling results, including BCL-2
dependence predicting treatment response. In addition, as our previous
work showed that ibrutinib could increase BCL-2 dependence, we further
hypothesize that CLL cells that are released into circulation would
still be exposed to the effects of ibrutinib to increase BCL-2
dependence. Taken together, the effects of ibrutinib treatment before
administering FCR would still allow us to properly predict treatment
responses via BH3-profiling. A final factor worth mentioning is the
translational implications of BH3-profiling as a tool for clinical
practice. Efforts are currently well underway to further standardize
the technique. Automated techniques have been employed pertaining to
preparation, handling and measurement, to minimize potential for human
error and facilitate possible future regulatory approval. We note that
BH3-profiling is particularly well-suited for circulating diseases such
as CLL due to easy sample availability and abundance from just a
routine blood draw. The very rapid turnaround time of just a few hours
and relatively low cost of the assay may facilitate uptake of the
technique in practice as additional validation studies continue to read
out.
Though our findings are robust, our study does have limitations to
highlight. For example, our validation cohort from the iFCR trial is a
population treated with a targeted therapy plus chemoimmunotherapy
combination. Future studies will be needed to validate whether our
results also apply in cohorts of patients treated with targeted
therapies alone. Additionally, the clinical data from both our
discovery and validation cohorts were not mature enough to analyze
time-to-event endpoints such as PFS and OS, which will be important to
evaluate in our future studies. Secondly, although our discovery cohort
of 73 patients encompasses various diverse patients characteristics
(i.e. genetic anomalies, prior treatment status, demographic diversity)
we acknowledge that there may still be potential selection biases in
populations of patients followed at tertiary academic hospitals
compared to patients followed in community practice. Finally, while a
group of untreated patients only was included in our iFCR cohort to
control for potential confounding effects from a mixed group of
previously treated or untreated patients, it would also be informative
to include patients that were previously treated in our future studies
to evaluate the potential confounding effects of prior treatments.
Conclusion
Unlike in many cancers, the novel agents approved in CLL do not target
recurrent mutations; rather, these new drugs targets pathways such as
BCR signaling and the BCL-2 family that are functionally important for
the survival of CLL cells. Thus, although genomic biomarkers will
continue to be an important way to understand the biology of CLL,
functional precision medicine strategies will also be crucial to help
optimize the efficacy of targeted agents in this disease. By directing
these agents preferentially to the patients most likely to benefit,
multi-dimensional functional and omics mapping will bring us closer to
the goal of personalizing therapy for patients with CLL. We identified
and defined BCL-2 dependence as a potential functional predictive
biomarker of conventional and targeted treatment response in CLL. This
underscores the importance of characterizing functional apoptotic
signaling in CLL to stratify patients beyond established genetic
biomarkers, which currently is less useful to guide treatment with
targeted agents, as well as to identify novel combinations to exploit
BCL-2 dependence therapeutically.
Supplementary Information
[238]Supplementary Material 1.^ (23KB, xlsx)
[239]Supplementary Material 2.^ (14KB, xlsx)
[240]Supplementary Material 3.^ (3.6MB, pdf)
[241]Supplementary Material 4.^ (821.9KB, xlsx)
[242]Supplementary Material 5.^ (19.8KB, docx)
Acknowledgements