IκB kinase‐α coordinates BRD4 and JAK/STAT signaling to subvert DNA damage‐based anticancer therapy

Abstract

Activation of the IκB kinase (IKK) complex has recurrently been linked to colorectal cancer (CRC) initiation and progression. However, identification of downstream effectors other than NF‐κB has remained elusive. Here, analysis of IKK‐dependent substrates in CRC cells after UV treatment revealed that phosphorylation of BRD4 by IKK‐α is required for its chromatin‐binding at target genes upon DNA damage. Moreover, IKK‐α induces the NF‐κB‐dependent transcription of the cytokine LIF, leading to STAT3 activation, association with BRD4 and recruitment to specific target genes. IKK‐α abrogation results in defective BRD4 and STAT3 functions and consequently irreparable DNA damage and apoptotic cell death upon different stimuli. Simultaneous inhibition of BRAF‐dependent IKK‐α activity, BRD4, and the JAK/STAT pathway enhanced the therapeutic potential of 5‐fluorouracil combined with irinotecan in CRC cells and is curative in a chemotherapy‐resistant xenograft model. Finally, coordinated expression of LIF and IKK‐α is a poor prognosis marker for CRC patients. Our data uncover a functional link between IKK‐α, BRD4, and JAK/STAT signaling with clinical relevance.

IKK‐α links apoptosis repression, DNA damage repair, and enhanced chemoresistance in colorectal cancer via targeting BET family protein BRD4 and NF‐κB/LIF‐induced STAT3.

Introduction

Colorectal cancer (CRC) accounts for about 12% of all deaths from cancer in the European countries (https://ec.europa.eu/eurostat/statistics‐explained/index.php/Cancer_statistics_‐_specific_cancers#Colorectal_cancer), which highlights the need for innovative therapeutic options. After years of research on cancer and cancer therapy, first‐line treatment for CRC is still surgery followed by radio‐ and/or chemotherapy for advance tumors. More recently, and based on the importance of mitogen‐activated protein kinase (MAPK) pathway in CRC, antibodies targeting the upstream regulator epidermal growth factor receptor (EGFR) have been included as a second therapeutic option. However, these therapeutic strategies are primarily restricted to the subset of tumors not carrying KRAS or BRAF mutations (Amado et al, 2008; Di Nicolantonio et al, 2008).

Nuclear factor kappa B (NF‐κB) signaling has been recurrently identified as a potent tumor driver and the essential linkage between inflammation and cancer. This circumstance led to investigate inhibitors of the Inhibitor of kappa B kinase (IKK) complex, the bottle neck of NF‐κB activation, as potential anticancer agents. However, this possibility was rapidly dismissed due to the extremely high toxicity of general NF‐κB inhibition (reviewed in Prescott & Cook, 2018), thus uncovering the need for identifying targetable elements of NF‐κB signaling specific of cancer. In this context, the IKK‐α kinase, which is dispensable for canonical NF‐κB signaling, has been the focus of intense investigation (reviewed in Colomer et al, 2017). Recently, we demonstrated that IKK‐α promotes therapy resistance in CRC by facilitating activation of the DNA damage repair (DDR) pathway through direct phosphorylation of ataxia telangiectasia mutated (ATM) kinase. Inhibition of IKK‐α activity by vemurafenib or AZ628 (Colomer et al, 2019) or IKK‐α genetic deletion precluded ATM‐dependent DDR following chemo‐ or radio‐therapy treatment leading to specific eradication of cancer cells in vitro and in vivo (Colomer et al, 2019).

Bromodomain 4 (BRD4) is a member of the bromodomains and extra‐terminal (BET) domain family of proteins, which facilitate polymerase II‐dependent transcription of genes through recognition of specific histone acetylation marks (Dey et al, 2003). BRD4 is unique among the human BET family proteins because it contains a carboxyl‐terminal domain (CTD) that interacts with the cyclin T1 and CDK9 subunits of positive transcription elongation factor b (pTEFb) complex to modulate polymerase II activity (Bisgrove et al, 2007). Moreover, BRD4 contains intrinsic histone acetyl transferase activity, which inflicts chromatin de‐compaction at its target genes (Devaiah et al, 2016). Additionally, BRD4 facilitates Homologous Recombination and Non‐Homologous‐End‐Joining DNA repair in cancer cells by direct regulation of DDR elements such as CHK1 (Li et al, 2018; Sun et al, 2018; Zhang et al, 2018; Wakita et al, 2020; Ni et al, 2021) while acting as chromatin insulator to limit the extent of ATM‐induced signaling after damage (Floyd et al, 2013).

BRD4 is functionally involved in cancer development (reviewed in Donati et al, 2018) and was found to impose a proliferative and anti‐apoptotic state to mouse embryonic fibroblasts (MEF) and cancer cell lines (Ray et al, 2014). BET inhibitors, such as JQ1, have reliably demonstrated their efficacy as anticancer agents and are currently under evaluation in several clinical trials (reviewed in Shorstova et al, 2021).

JAK/STAT pathway is a well‐known tumor driver in different cancer subtypes (reviewed in Yu et al, 2014a), which promotes cell proliferation and therapy resistance by different mechanisms including transcriptional activation of MYC (Bowman et al, 2001), CYCLIN D1 (Leslie et al, 2006), MDM2 (Yu et al, 2014b), and several antiapoptotic genes (reviewed in Jones et al, 2016). JAK/STAT signaling is induced by growth factors and cytokines such as interleukin 6 (IL6) or the leukemia inhibitory factor (LIF; Hu et al, 2021). JAK‐mediated tyrosine (Y) phosphorylation of STAT (Y705 for STAT3) leads to STAT dimerization, nuclear translocation, and specific gene transcription (Darnell et al, 1994). LIF has been described to promote the development and progression of numerous solid tumor types (Yu et al, 2014b; Liu et al, 2019; McLean et al, 2019). It can induce the self‐renewal of cancer stem cells/cancer‐initiating cells (Peñuelas et al, 2009) and promote immune suppression in tumors (Pascual‐García et al, 2019; Hallett et al, 2023).

We now show that IKK‐α coordinates BRD4 and STAT3 functions through (i) direct phosphorylation of BRD4 thus regulating chromatin‐binding dynamics and (ii) NF‐κB‐mediated transcriptional induction of LIF leading to STAT3 activation, formation of BRD4 and STAT3 complexes, and BRD4 recruitment to a subset of STAT3 target genes. Simultaneous inhibition of JAK/STAT and nuclear IKK‐α activity with BRAF inhibitors (Colomer et al, 2019) or BRD4 with JQ1 enhanced the anticancer activity of chemotherapy in human CRC cells and tumor xenografts. We identify LIF together with IKK‐α as a prognosis marker for CRC patients.

Together, our data offer new insights on IKK‐α function in normal and tumor cells and provide novel targets for personalized anticancer therapies.

Results

BRD4 S1117 is a phosphorylation substrate of the IKKα kinase

By mass spectrometry (MS) analysis of control and IKK‐α knocked‐down HT29 CRC cells treated with UV (Colomer et al, 2019), we identified serine (S) 1117 of BRD4 as a UV‐inducible and IKK‐α‐dependent phosphorylated residue (Fig 1A). This observation raised the possibility that IKK‐α directly phosphorylates BRD4. By in vitro kinase assay, we confirmed that recombinant active IKK‐α was able to phosphorylate a BRD4 fragment comprising amino acids 1,037–1,201 (Fig 1B). Mutation of S1117 into alanine precluded BRD4 phosphorylation induced by IKK‐α (Fig 1C). We identified a second BRD4 residue (S601) whose phosphorylation increased after UV treatment independently of the IKK‐α status (Fig 1D). Accordingly, recombinant IKK‐α failed to phosphorylate two different BRD4 fragments containing S601 (Fig 1E).

Figure 1. BRD4 S1117 is a phosphorylation substrate of IKK.

• A
  Representation of the area corresponding to the phosphor‐peptide containing S1117 of IKK‐α as determined by mass spectrometry analysis (MS) of control and IKK‐α knocked‐down HT29 cells untreated or exposed to UV light (130 mJ) for 30 min (n = 5 (untreated) and 2 (treated) biologically independent replicates).
• B, C
  In vitro kinase assay with recombinant IKK‐α and purified glutathione S‐transferase (GST), GST‐BRD4 (amino acids [aa] 1,037–1,201) (B) or the same GST‐BRD4 fragment including a S > A mutation at residue 1117 (C) (from one out of three independent experiments).
• D
  Representation of the area corresponding to the phosphor‐peptide including S601 from BRD4 as determined by mass spectrometry (MS) analysis of control and IKKα‐knock‐down HT29 cells untreated or exposed to UV light (130 mJ) for 30 min (n = 3 biologically independent replicates).
• E
  In vitro kinase assay with recombinant IKK‐α and purified GST, two different GST‐BRD4 fragments including S601 of BRD4 or GST‐ BRD4 S1117 [aa1037‐1201] as positive control (from one out of three independent experiments).

Data information: Bars in A and D represent mean values and standard error of the mean (s.e.m.); P‐values were derived from one‐way ANOVA. ***P‐value < 0.001; n.s.: no significant.

Source data are available online for this figure.

These results indicate that IKK‐α specifically phosphorylates BRD4 at S1117. The high conservation of this residue among species (Fig EV1) further suggested its functional relevance.

Figure EV1. BRD4 S1117 is a phosphorylation substrate of the IKK‐α kinase.

Residue conservation of position S1117A is highlighted in yellow in the alignment of the BRD4 proteins.

Two different domains of BRD4 mediate binding to IKKα

By immunoprecipitation (IP) assay of UV‐treated cancer cells with the antibody targeting IKK‐α (p45), we demonstrated the physical binding of IKK‐α to BRD4 that was slightly increased at 5 min of UV exposure and maintained after 15 min (Fig 2A). The inducible nature of the IKK‐α and BRD4 interaction was confirmed by BioID assay using an IKK‐α protein fused to a biotin donor (see methods). Specifically, upon association with Bio‐IKK‐α, BRD4 become biotinylated and is subsequently recovered in streptavidin precipitates. We observed a significant enrichment in the amount of biotinylated BRD4 after 15 min of UV treatment of Bio‐IKK‐α‐expressing HCT116 cells, which decreased at later time points (i.e. 60 min; Fig 2B). We further confirmed the IKK‐α and BRD4 interaction by reciprocal IP with the BRD4 antibody (Fig 2C).

Figure 2. Two different domains of BRD4 mediate IKK‐α binding.

1. Immunoprecipitation assay (IP) with anti‐IKKα (p45) antibody from CRC HT29 cells exposed to UV light (130 mJ) and collected at the indicated time points (from one out of three independent experiments).
2. Cells lysates from HCT116 cells expressing a fusion BioID‐IKK‐α protein were precipitated using streptavidin‐coated beads and then analyzed by western blot with the indicated antibodies (from one out of three independent experiments).
3. IP assay with anti‐BRD4 antibody from HT29 cell line untreated or collected 15 min after UV light exposure (130 mJ) (from one out of three independent experiments).
4. Pull‐down assay (PD) of cell lysates from HT29 cell line untreated or collected 15 min after UV light exposure (130 mJ) using GST fused to the indicated fragments of BRD4 as bait. Schematic representation of the fragments used for mapping the domains of BRD4 involved in IKKα binding (IKK‐α Binding Domains 1 and 2) (from one out of three independent experiments).

Data information: Inputs in A, B and C correspond to 1/25 of the precipitated lysate.

Source data are available online for this figure.

We next mapped the interaction domain of BRD4 to IKK‐α by pull‐down assay using various GST/BRD4 fusion protein fragments. Our results indicated that IKK‐α binds to two different regions of BRD4, the first one involving amino acids 301 to 320 and the second one corresponding to the carboxyterminal domain (CTD), which is important for BRD4‐mediated transcriptional activation (Jang et al, 2005) and includes the IKK‐α phosphorylation residue 1,117 (Fig 2D). Interestingly, activated IKK‐α (recovered 15 min after UV treatment) failed to stably associate with the CTD fragment (Fig 2D) consistent with the transient binding of a kinase with its substrate, which may not necessarily represent the actual binding dynamics between IKK‐α and BRD4 in vivo.

IKK‐α‐induced phosphorylation regulates BRD4 chromatin‐binding dynamics and impacts on activation of the DDR elements ATM and Chk1

Association of BRD4 with the chromatin follows a cycling pattern after exposure to DNA damaging agents, being chromatin dissociation and reassociation essential for BRD4‐mediated gene transcription (Ai et al, 2011). We confirmed the cycling nature of BRD4 chromatin binding upon exposure to different damaging agents including UV light, ionizing radiation (IR), and chemotherapy (5‐FU + Iri) in different cell lines (Figs 3A and B, and EV2A–E). Chromatin dissociation and reassociation of BRD4 display different dynamics in the several models tested, and it was severely compromised in IKK‐α‐deficient cells (Fig EV2A and B) and in cells treated with the BRAF inhibitor AZ628 (Fig 3A), which prevents IKK‐α (p45) activation by damage (Colomer et al, 2019). However, histone acetylation, that is one of the factors regulating BRD4 chromatin binding, was not significantly affected by IKK‐α deletion or BRAF inhibition (Figs 3A and EV2A–C, E, and F). To test the impact of IKK‐α‐induced BRD4 phosphorylation in chromatin‐binding dynamics, we generated knock‐in HCT116 cells carrying a BRD4 mutant in which S1117 has been changed to A (BRD4_S1117A, see methods). BRD4_S1117A mutation imposed a massive accumulation of BRD4 in the soluble nuclear compartment and impaired chromatin‐binding dynamics when compared with WT BRD4 (Fig 3B). Remarkably, BRD4_S1117A HCT116 cells showed defective activation of the DDR elements ATM, ATR, and CHK1 upon UV treatment, and increased apoptosis as determined by cleaved caspase 3 levels (Fig 3C), a phenotype that paralleled the effect of full BRD4 inhibition by JQ1 treatment in these cells (Fig EV2G). These results suggest that IKK‐α‐mediated phosphorylation of BRD4 at S1117 is required for its correct chromatin‐binding dynamics and function, including DDR. By qPCR analysis, we found that expression of several DDR‐related elements remained unaffected, or even increased, in cells carrying the BRD4_S1117A mutant (Fig 3D), supporting the concept that BRD4 function on DDR is gene expression independent (Li et al, 2018; Barrows et al, 2022) but dependent on IKK‐α‐induced phosphorylation.

Figure 3. IKK‐α‐induced phosphorylation regulates BRD4 chromatin‐binding dynamics.

1. WB analysis of chromatin extracts from HT29 cell line treated with BRAF inhibitor (AZ628, 10 nM) for 16 h before UV light exposure (130 mJ) and collected as indicated time points (from one out of three independent experiments).
2. HCT116 BRD4 WT and HCT116 BRD4_S1117A knock‐in cells exposed to UV light (130 mJ) at the indicated time points followed by subcellular fractionation (cytoplasm, nucleus and chromatin) and WB analysis (from one out of three independent experiments).
3. Western blot analysis of HCT116 BRD4 WT and HCT116 BRD4_S1117A knock‐in cells treated as indicated (from one out of three independent experiments).
4. Expression analysis by qPCR of the indicated DDR‐related genes from WT and BRD4_S1117A knock‐in HCT116 cells. Bars represent mean values and standard error of the mean (s.e.m.) from six independent replicates; P‐values were derived from two‐sided unpaired t‐test. ****P‐value < 0.0001, **P‐value < 0.01.

Source data are available online for this figure.

Figure EV2. BRD4 chromatin dynamics after DNA damage.

• A, B
  Western blot analysis (WB) of chromatin extracts from MEFs (A) and Caco2 (B) IKK‐α WT and IKK‐α KO exposed to UV light (130 mJ) and collected at the indicated time points (from one out of three independent experiments).
• C, D
  WB analysis of cytoplasmic, nuclear, and chromatin extract from HT29 cell line collected at different time points after UV treatment (130 mJ) (A) or treated with 5 FU + Irinotecan (5FU = 5 μg/ml, Irinotecan = 2 μg/ml) at the indicated time points (B) (from one out of three independent experiments).
• E
  WB analysis of chromatin extract from HT29 cells collected at different time points after IR exposure (10Gy) (from one out of three independent experiments).
• F
  Densitometric analysis of the ratio of acetylated K5 of histone H4 and total H4 from Figs 3A (upper graph), 3B (medium graph), and 3C (lower graph). Since the sum of relative acK5 in WT and KO (or AZ628‐treated) cells was adjusted a 100 in the graphs, the confluence of orange and blue bars at 50 (dotted line) indicates identical acetylation levels in both systems.
• G
  WB analysis of HCT116 cells untreated or pretreated with the BRD4 inhibitor JQ1 for 16 h and the exposed to UV as indicated (from one out of three independent experiments).

IKK‐α dictates BRD4 chromatin binding to specific STAT3 target genes

To investigate whether aberrant chromatin‐binding dynamics in IKKα‐deficient cells led to different genomic BRD4 occupancy after damage, we took advantage of the murine organoid model derived from APC^Min/+ IKKα WT and KO intestinal tumors (Colomer et al, 2018). We first confirmed that S1117 of BRD4 was phosphorylated in APC^Min/+ intestinal organoids (Fig EV3A) and IKK‐α deficiency precluded chromatin‐binding dynamics of BRD4 upon IR (Fig EV3B). We performed ChIP assay with BRD4 antibody from IKK‐α WT and KO APC^Min/+ organoids left untreated or treated with IR and recovered after 60 min. We observed a comparable distribution of BRD4 at the different genomic regions (intergenic, promoter, or gene body) between IKK‐α WT and KO cells both in untreated conditions and after IR (Fig EV3C). The read coverage extracted from ChIP‐seq experiments across 1 kb non‐overlapping genomic windows showed a significant correlation between WT and IKK‐α KO cells in non‐irradiated conditions (Fig 4A), indicating that basal BRD4 binding landscape was IKK‐α independent. However, IR treatment resulted in the detection in IKK‐α WT cells of a subset of BRD4‐bound regions corresponding to 774 genes (Fig 4B, red dots) that were marginally detected in IKK‐α‐deficient cells (Fig 4C, purple dots). These results indicated that a number of IR‐induced BRD4 binding sites were IKK‐α dependent, which was further supported by correlation analysis of IR‐treated IKK‐α WT and KO cells (purple dots in Fig 4D). Next, we performed a transcription factor (TF) target gene enrichment analysis of regions bound by BRD4 in the different experimental conditions. In non‐irradiated conditions, BRD4‐bound regions showed a significant representation of target genes for multiple transcription factors including MYC, UBTF, GATA2, or RELA (Fig 4E). Interestingly, BRD4‐bound genes exclusively detected in IR‐treated IKK‐α WT cells were specifically enriched in UBTF, TRIM28, and STAT3 target genes (Fig 4E). We observed a significant correlation between IKK‐α‐ and IR‐dependent BRD4‐bound genes (Fig 4F) with STAT3 target genes (Carpenter & Lo, 2014; Robinson et al, 2019; Fig 4G and Dataset EV1). Among the STAT3 target genes recruiting BRD4 in an IKK‐α‐dependent manner, we identified Bcl3 (Brocke‐Heidrich et al, 2006; Dataset EV1 and Fig EV3D), which is an essential inhibitor of apoptosis in several systems (Chang & Vancurova, 2014; Urban et al, 2016) and facilitates immune scape in cancer through PDL1 upregulation (Zou et al, 2018). By qPCR, we found that basal and IR‐induced Bcl3 expression was decreased in IKK‐α‐deficient APC^Min/+ organoids (Fig 4H) and MEF cells (Fig 4I). Bcl3 levels were also reduced in cells treated with the BRD4 inhibitor JQ1 (Fig 4J) and the JAK/STAT inhibitor ruxolitinib (Fig 4K), indicating that Bcl3 expression requires simultaneous BRD4 and STAT3 activity. We confirmed altered IR‐induced expression of several double BRD4 and STAT3 targets by qPCR analysis of APC^Min/+ IKK‐α KO organoids (Fig EV3E).

Figure EV3. IKK‐α dictates BRD4 chromatin recruitment upon damage.

1. Graphical representation of the area corresponding to the BRD4 peptide including the phosphorylated S1117 present in the lysates from APC^Min/+‐derived organoids.
2. WB analysis of chromatin lysates from APC^Min/+‐derived organoids exposed to IR and recovered at the indicated time points (from one out of three independent experiments).
3. Genomic distribution of the different BRD4‐enriched regions. Random: randomly selected regions of the genome; BRD4_all: BRD4‐bound regions regardless IKK‐α genotype and treatment conditions; CWT: BRD4‐bound regions in untreated WT cells, WTIR_KOIR: BRD4‐bound upon treatment independent of IKK‐α genotype; WTIR: BRD4‐bound regions upon IR in WT and, KOIR: BRD4‐bound upon IR in IKK‐α KO.
4. Representation of BRD4 distribution (from ChIP sequencing) in the BCL3 gene promoter in untreated or IR‐treated WT and IKKα KO APC^Min/+ organoids. The promoter region that is specifically bound by BRD4 in IR‐treated IKK‐α WT cells is outlined in the figure.
5. qPCR analysis of the indicated genes in WT and IKK‐α KO APC^Min/+ organoids treated with IR for 30 min. Bars represent mean values and standard error of the mean (s.e.m.) from two independent replicates; P‐values were derived from two‐sided unpaired t‐test.

Figure 4. IKK‐α dictates BRD4 chromatin binding to specific STAT3 target genes.

• A–D
  Correlation of ChIP‐seq depth across the genome of untreated or IR‐treated WT and IKK‐α KO APC^Min/+ organoid samples. Untreated IKK‐α WT and KO (A), IR‐treated and untreated IKK‐α WT (B), IR‐treated and untreated IKK‐α KO (C) and IR‐treated IKK‐α WT and KO (D). In all the plots dots correspond to 1 kb non‐overlapping genome windows. Dots in blue represent genomic regions that show a higher coverage in untreated WT compared with IR‐treated WT, according to a linear model. In red are regions that show a higher coverage in IR‐treated WT compared with untreated WT. In purple, regions with a higher coverage in IR‐treated KO compared to untreated KO; and in yellow, regions with a higher coverage in IR‐treated IKK‐α KO compared to IR‐treated WT.
• E
  Enrichment analysis of different TFs targets from ENCODE and CHEA databases within each group of BRD4‐bound regions (CWT, regions specifically detected in the WT untreated; CKO, regions detected in untreated IKK‐α KO; “random” refers to randomly selected regions of the genome; WTIR, regions exclusively detected in IR‐treated IKK‐α WT; and KOIR regions exclusively found in IR‐treated IKK‐α KO cells).
• F, G
  Venn diagrams representing the number of BRD4‐bound genes upon IR treatment that are IKK‐α dependent (F) and those who have been previously identified as STAT3 targets (G).
• H–K
  qPCR gene expression analysis of the double BRD4 and STAT3 target gene Bcl3 in APC^Min/+ organoids (H, J and K) or MEFs (I) treated as indicated (IR; 10 Gy). Reduced levels of MYC (J) and p‐STAT3 (K) indicated the efficacy of the treatments. Bars in (H–K) represent mean values from two independent replicates; P‐values were derived from two‐sided unpaired t‐test in (H), (I) and (K) and from one‐way ANOVA in (J). **P‐value < 0.01, *P‐value < 0.05; n.s.: no significant.
• L, M
  IP with anti‐BRD4 antibody from untreated or UV‐treated IKK‐α WT and KO APC^Min/+ organoids (L) or IKK‐α WT MEFs untreated or treated with 10 μM ruxolitinib (Rux) for 16 h and then exposed to UV light (130 mJ) for 15 min (M). (one out of three independent experiments is shown).
• N
  IP with anti‐BRD4 antibody from parental and HCT116 S1117A knock‐in HCT116 cells collected 15 min. after UV exposure (one out of three independent experiments performed id shown).

Data information: In L–N, inputs correspond to 1/25 of the precipitated lysate.

Source data are available online for this figure.

These results suggested that IKK‐α was coordinating BRD4‐ and STAT3‐dependent transcription upon damage. Supporting this concept, we detected basal BRD4 and STAT3 interaction by CoIP, which was increased after UV treatment and abrogated in IKKα KO cells (Fig 4L). Basal and UV‐induced association between BRD4 with STAT3 was abolished by the JAK/STAT inhibitor ruxolitinib (Fig 4M) but unaffected by the BRD4 S1117 mutation (Fig 4N). Together, these results indicate that BRD4 preferentially binds to active STAT3, which is IKK‐α dependent. However, BRD4 and STAT3 interaction is independent of BRD4 phosphorylation by IKK‐α.

IKK‐α deficiency precludes STAT3 activation in the presence of a functional JAK/STAT signalosome

We studied the possibility that IKK was regulating STAT3 activation upon damage. In agreement with this possibility, we detected basal and damage‐induced phosphorylation of STAT3 at tyrosine 705 (target of the Janus kinase, JAK, and marker of canonical STAT3 activation) in IKK‐α WT cells, which was primarily abolished in cells lacking IKK‐α (Figs 5A and EV4A–E). However, STAT3 activation was only marginally decreased after inhibition of nuclear IKK‐α (p45) by AZ628 (Margalef et al, 2015; Colomer et al, 2019; Fig EV4F), suggesting that was canonical IKK activity that regulates JAK/STAT. Since IKKs does not possess tyrosine kinase activity, we reasoned that IKK‐α may regulate the expression or activity of one or more upstream elements of the JAK/STAT signaling pathway. By qPCR analysis, we found that mRNA levels of JAK1, JAK2, IL6R, LIFR, and the GP130 co‐receptor were comparable or higher in IKK‐α KO MEFs than in WT cells (Fig 5B). Supporting the functional integrity of the JAK/STAT signalosome, levels of p‐STAT3(Y705) were robustly induced by treating IKK‐α KO MEFs with the canonical JAK/STAT activators IL6 and leukemia inhibitory factor (LIF; Fig 5C). Conditioned media (CM) from UV‐treated WT MEFs also induced STAT3 activation in IKK‐α KO MEFs at levels comparable to IL6 or LIF (Fig 5C). Moreover, treatment of IKK‐α KO MEFs with LIF was sufficient to induce BCL3 transcription, which was not observed upon IL6 treatment (Fig 5D).

Figure 5. IKK‐α regulates STAT3 signaling by its paracrine activation through LIF.

1. WB analysis of protein extracts from WT and IKK‐α KO APC^Min/+ organoids exposed to ionizing radiation (IR) (10 Gy) and collected at the indicated time points (one out of three independent experiments is shown).
2. qPCR analysis of the indicated genes in MEFs IKK‐α WT and IKK‐α KO.
3. WB analysis of extracts from MEFs IKK‐α WT and KO pretreated with IL‐6 (50 ng/ml), IL‐3 (100 μg/ml), LIF (10 ng/ml) (Leukemia inhibitory factor) or ruxolitinib (Rux, 2.5 μM) for 16 h, or conditioned media (C.M.) from UV‐treated (30 min.) WT MEFs (one out of two independent experiments performed is shown).
4. qPCR analysis of WT and IKK‐α KO cells untreated or treated with IL‐6 (50 ng/ml) or LIF (10 ng/ml) for 16 h.

Data information: Bars in (D) and (F) represent the mean value and standard error of the mean (s.e.m.) of three and two independent replicates performed, respectively. P‐values were derived from two‐sided unpaired t‐test in (D) and from one‐way ANOVA in (F). ****P‐value < 0.0001, *P‐value < 0.05; n.s.: no significant.

Source data are available online for this figure.

Figure EV4. IKK‐α facilitates STAT3 phosphorylation and transcriptional activity after damage.

• A–E
  WB analysis of WT and IKKα KO MEFs (A), Caco2 (B), Hela (C, D), and PDO5 cells (E) treated as indicated (one out of three independent experiments is shown).
• F
  WB analysis of HT29 CRC cells untreated or pretreated with the BRAF inhibitor AZ628 for 16 h and then exposed to IR as indicated (from one out of three independent experiments).

Together, these results indicate that IKK‐α contributes to STAT3 activity through the regulation of one or more secreted JAK/STAT activating factors.

JAK/STAT and IKK‐α (p45) inhibition potentiates chemotherapy in CRC patient‐derived organoids in vitro

We previously demonstrated that IKK‐α provides therapy resistance to CRC PDO cells (Colomer et al, 2019). We investigated whether IKK‐α‐mediated regulation of the apoptosis inhibitor BCL3 (through BRD4 and STAT3) contributes to this effect. By WB analysis of cleaved‐PARP1 demonstrated that IKK‐α deficiency increases apoptosis of PDO5 (Fig 6A) and PDO8 (Fig EV5A) cells upon chemotherapy treatment. Higher chemotherapy sensitivity of PDO5 cells was also observed upon treatment with the JAK/STAT inhibitor ruxolitinib or JQ1 (P < 0.0001, for both conditions, two‐way ANOVA test; Fig 6B). Inhibition of IKK‐α (p45) by AZ628, that has little impact on STAT3 activation (Fig EV4F), increased CT‐induced cell death, which was further enhanced by ruxolitinib addition (Figs 6C and EV5B; 5‐FU + Iri, IC50 = 2.37 mg/ml; 5FU + Iri + AZ628, IC50 = 0.025 mg/ml, and IC50 = 1.982e‐006 mg/ml for the quadruple combination in PDO5. P < 0.0001 for all comparisons, two‐way ANOVA test). In contrast, we did not detect any additional benefit of combining AZ628 with the BRD4 inhibitor JQ1 (Fig EV5C) in agreement with the concept that AZ628 already inhibits IKK‐α‐dependent BRD4 activity in CRC cells (see Fig 3A).

Figure 6. Addition of JAK/STAT and BRAF inhibitors to therapeutic treatment leads to effective eradication of chemotherapy‐resistant human tumors.

• A
  WB analysis of IKK‐α WT and KO PDO5 cells treated with IC20 5‐FU + Iri for 72 h and ruxolitinib (Rux, 40 μM) for 16 h (from one out of two independent experiments).
• B, C
  Dose–response curves of PDO5 cells treated with 5‐FU + Iri alone or in combination with ruxolitinib [40 μM] (B) or 5‐FU + Iri and AZ628 [25 μM] with or without ruxolitinib (C) for 72 h (one out of three biologically replicates performed is shown). 5‐FU, 5‐fluorouracil: Iri, irinotecan.
• D–F
  Photograph of tumors recovered from mice with xenografts in the indicated groups of treatment (D) and quantification of the size (E) and weight (F) of tumors. Of note that two of the tumors in the vehicle group were sacrificed during the weekend and processed by the personal in the animal facility, consequently they were not photographed and only one was measured.
• G, H
  Representative microscopic images of tumors recovered from mice in the indicated groups of treatment showing the areas containing alive tumor cells labeled as T and delimited by dashed lines (G). Detail of the residual tumor areas (H). Notice the presence of massive fibrotic and necrotic areas surrounding the tumor areas. Number of tumors displaying the exposed phenotype (X) from the total of animals in each group (N) is indicated in the upper‐right corner of the images (X/N).
• I, J
  IHC analysis of the proliferation marker ki67 in the remaining tumor areas of the different groups of treatment (I) and quantification (J) of 10 areas per group (40X), when possible.
• K
  Kaplan–Meier curves for mice bearing a metastatic human tumor treated as indicated. Overall survival is indicated for the different treatments and statistical significance was determined using the Mantel‐Cox log‐rank test. In the legends Vem, vemurafenib; 5‐FU, 5‐fluorouracil; Rux, ruxolitinib and Iri, irinotecan.
• L, M
  Total weight (L) and volume (M) of tumors recovered at humanitarian endpoint for each mouse or at the end of the experiment in the case of quadruple combination treatment group.

Data information: Bars in E, F, J, L and M represent mean values and standard deviation and statistical significance was determined by one‐way ANOVA test and the Tukey's Multiple Comparison Analysis. *P‐value < 0.05, **P‐value < 0.01, ***P‐value < 0.001, ****P‐value < 0.0001; n.s.: no significant.

Source data are available online for this figure.

Figure EV5. Increased cell death of IKK‐α KO cells in response to CT. Association of IKK‐α with LIF.

• A
  WB analysis of IKK‐α WT and KO PDO8 cells treated with IC20 5‐FU + Iri for 72 h and with ruxolitinib for 16 h (from one out of two independent experiments).
• B, C
  Dose–response assay of PDO4 (G) and PDO5 cells (H) treated as indicated. In all the experiments, the vehicle of the inhibitors was added to the 5‐FU + Iri only treatment (n = 3 replicates examined, from one out of two biologically independent experiments). 5‐FU, 5‐fluorouracil: Iri, irinotecan.
• D
  Representative image of ki67 staining in the normal colonic mucosa adjacent to the tumors treated with the triple combination 5‐FU + Iri., vemurafenib and ruxolitinib.
• E–H
  Photograph of tumors recovered from mice in the indicated groups of treatment (E) and quantification of the size (F), weight (G), and percent of proliferating (ki67⁺) cells (H) in the different tumors. Bars in (F–H) represent mean values and standard deviation. Statistical significance was determined by one‐way ANOVA test and the Tukey's multiple comparison analysis. *P‐value < 0.05, **P‐value < 0.01, ***P‐value < 0.001, ****P‐value < 0.0001; n.s.; no significant.
• I
  qPCR analysis of LIF, LIFR, and GP130 levels in the indicated cellular models (n = 2 (HT29, Caco2, HeLa, PDO8, MEFs), 3 (HCT116, PDO5), and 4 (APCmin) biologically independent experiments).
• J
  Graphical representation of LIF mRNA levels in the CRC tumors deposited at the TCGA dataset.
• K
  Violin plots showing LIF mRNA levels in the CRC tumors classified as CHUK high and CHUK low in the TCGA dataset. Boxes represent the central 50% of the data (from the lower 25^th percentile to the upper 75^th percentile), lines inside boxes represent the median (50^th percentile), and whiskers are extended to the most extreme data point. Statistical P‐value from Wilcoxon two‐sided test is shown.
• L, M
  Spearman's rank correlation analysis of LIF and CHUK levels in human CRC tumors from the TCGA dataset including all stages (E) (n = 329) or considering stage II tumors only (F) (n = 106).

JAK/STAT and IKK‐α (p45) inhibition together with chemotherapy allows effective eradication of CRC xenografts in vivo

To test the therapeutic value of STAT3 and IKK‐α (p45) inhibition in vivo, we implanted equivalent volumes of a human KRAS‐mutated CRC tumor with partial resistance to 5‐FU + Iri (CRC#3) in the cecum of athymic nude mice. Tumor growth was monitored by palpation and, at the time of tumor detection, mice were randomly ascribed to the treatment groups: Vehicle, 5‐FU + Iri, 5‐FU + Iri plus the BRAF inhibitor vemurafenib (Vem.), 5‐FU + Iri plus ruxolitinib (Rux) or the quadruple combination 5‐FU + Iri + Vem + Rux. Combination of 5‐FU + Iri plus vemurafenib significantly reduced tumor growth, as previously demonstrated (Colomer et al, 2019), which was only slightly improved in the quadruple treatment group (Fig 6D–F). Microscopic examination of the tumors demonstrated the presence of extensive areas of necrosis and fibrosis in tumors treated with different drug combinations. Notably, these areas practically occupied the whole tumor mass in the quadruple treatment group (Fig 6G, see pink areas surrounding the alive tumor territory, T), with residual neoplastic cells displaying a severe pleomorphism associated with the presence of pyknotic nuclei (Fig 6H). We detected a substantial inhibition of cell proliferation in the residual tumor areas of all tested combination treatments, which was more significant in the quadruple combination group (Fig 6I and J). Importantly, quadruple combination treatment did not impose general toxicity in the animals as indicated by the overall appearance of mice and the absence of histological or functional alterations (i.e. proliferation) in the colonic tissue adjacent to the tumors (Fig EV5D). We obtained comparable results in a second in vivo experiment using an APC, KRAS, and PI3K mutant CRC pulmonary metastasis (MPL‐CRC7; Figs EV5E–H).

Then, we performed a long‐term in vivo assay using a third CRT‐resistant CRC pulmonary metastasis carrying APC, KRAS, and TP53 mutations (MPL‐CRC4). In agreement with the chemotherapy resistance found in the patient, we did not observe any survival improvement in the 5‐FU + Iri‐treated group compared with the control, which was clearly observed upon addition of vemurafenib, ruxolitinib or both. Importantly, none of the animals treated with the quadruple combination succumbed in the period of the study (Fig 6K). Tumors recovered from moribund animals did not show significant differences in weight or volume, independently of the treatment. However, tumors present in the quadruple treatment group at the experimental endpoint were largely reduced compared with all other groups, even when the latter had been sacrificed earlier (Fig 6L and M). These results indicate that combination of BRAF and JAK/STAT inhibitors enhances the long‐term therapeutic benefit of chemotherapy on tumors with different mutational burdens, including those that have acquired resistance to first‐line chemotherapy treatment.

LIF mediates JAK/STAT activation by IKK‐α thus imposing therapeutic resistance in CRC cells

We have previously identified the leukemia inhibitory factor (LIF) as a crucial mediator of JAK/STAT activation (Peñuelas et al, 2009) and immunotherapy refraction (Pascual‐García et al, 2019) in cancer. We investigated whether LIF‐induced JAK/STAT signaling downstream of IKK‐α. By qPCR analysis, we detected variable levels of LIF, LIFR, and GP130 mRNA in all tested cellular models (Fig EV5I), which were consistently reduced following IKK‐α depletion (Fig 7A). By ChIP assay, we detected p65/NF‐κB binding to the promoter region of the murine LIF gene specifically in the IKK‐α WT cells (Fig 7B), indicating that LIF transcription was induced by canonical NF‐κB, downstream of IKK‐α. Abrogation of LIF activity with the anti‐LIF blocking antibody (Fig 7C) or with two specific shRNAs (Fig 7D) led to a reduction in p‐STAT3 (Y705) levels in IKK‐α WT MEFs and precluded the capacity of conditioned media to induce STAT3 activation in IKKα KO cells (Fig 7C and D).

Figure 7. LIF mediates JAK/STAT activation by IKK‐α thus imposing therapeutic resistance in CRC cells and poor patient outcome.

• A
  qPCR analysis of LIF levels in the indicated IKK‐α WT and KO cells (n = 2 independent replicates for APC^Min and n = 3 for MEFs, PDO5 and Caco2).
• B
  ChIP‐qPCR analysis of p65/NF‐κB binding to the promoter region of the murine LIF gene in MEFs IKK‐α WT and KO (n = 3 biologically replicates performed).
• C
  WB analysis of untreated and anti‐LIF‐treated (100 μg/ml for 2 h) IKK‐α WT MEFs and IKK‐α KO MEFs incubated for 2 h with control or anti‐LIF‐treated IKK‐α WT conditioned media (C.M.) (one out of two independent experiments is shown).
• D
  WB analysis (left panel) of control and LIF shRNA‐transduced IKKα WT MEFs and IKK‐α KO incubated with control medium or conditioned media from the indicated cells treated with IR for 30 min (from one out of two independent experiments). qPCR analysis of LIF levels (right panel) of control and LIF shRNA‐transduced IKK‐α WT and IKKα KO MEFs (n = 2 independent experiments performed).
• E, F
  Survival assay in HCT116 CRC cells treated as indicated (E) and WB analysis of the same cells with the indicated antibodies (F) to evaluate the effects of the α‐LIF antibody and JQ1 on p‐STAT3 and MYC levels, respectively (n = 3 biologically replicated performed).
• G
  Box plots of LIF mRNA levels in tumors samples (n = 566) compared with normal samples (n = 19) from the Marisa dataset. Boxes represent the central 50% of the data (from the lower 25^th percentile to the upper 75^th percentile), lines inside boxes represent the median (50^th percentile), and whiskers are extended to the most extreme data point. Statistical P‐value from Wilcoxon two‐sided test is shown.
• H
  Barplot depicting the normalized enrichment score of statistically significant enriched pathways obtained by GSEA analysis with the Hallmark gene set for patients with high‐LIF expression (NOM P‐value < 0.05).
• I, J
  GSEA of the TNFα via NF‐κB (B) and JAK/STAT (C) pathway in high‐LIF (n = 159) versus low LIF (n = 170) tumors in TCGA patients. P‐value is nominal P‐value given by the GSEA program in (J) and (K).
• K
  Pie charts showing the molecular subtype distribution according to Guinney et al (Guinney et al, 2015) in patients carrying LIF high (n = 55) and LIF low (n = 51) tumors from the stage II TCGA cohort. From 55 high‐LIF tumors, 41 with CMS information: CMS1 n = 16, CMS2 n = 12, CMS3 n = 4 and CMS4 n = 9. From 51 low LIF tumors, 41 with CMS information: CMS1 n = 4, CMS2 n = 13, CMS3 n = 9 and CMS4 n = 15.
• L, M
  Kaplan–Meier representation of disease‐free survival (DFS) over time for all patients in the TCGA CRC cohort (E) (LIF high n = 222 and LIF low n = 107) and selected stage II patients (F) (LIG high n = 37 and LIF low n = 37) according to LIF mRNA levels (optimized cutoff of 8.94 and 9.89, respectively).
• N, O
  Kaplan–Meier representation of disease‐free survival (DFS) over time for stage II patients in the indicated groups according to CHUK and LIF levels.

Data information: In A, B, D and E, bars represent mean values and standard error of the mean (s.e.m.). P‐values were derived from two‐sided unpaired t‐test in (A) and B and from one‐way ANOVA in (D) and (E). ****P‐value < 0.0001, ***P‐value < 0.001, **P‐value < 0.01, *P‐value < 0.05; n.s.: no significant. In L‐O, the optimized cutoff for LIF high and low was determined using the ‘maxstat’ R package, and we used Cox proportional hazards models for statistical Kaplan–Meier analysis and log‐rank two‐sided P‐value. HR, hazard ratio.

Source data are available online for this figure.

We then tested the therapeutic potential of combining LIF and BRD4 inhibition with chemotherapy in CRC cells. We treated HCT116 cells with sublethal 5‐FU + Iri alone or in combination with JQ1, anti‐LIF blocking antibody or both. Adding JQ1 or anti‐LIF antibody to the chemotherapy treatment did not improve chemotherapy efficacy, which was significantly observed in the triple combination (Fig 7E). By WB assay, we found that anti‐LIF and JQ1 treatments led a significant reduction in the levels of p‐STAT3 and the BRD4 target c‐MYC (Delmore et al, 2011), respectively (Fig 7F).

LIF and CHUK levels identify poor outcome CRC patients

Finally, we investigated whether the IKK‐α‐LIF‐JAK/STAT axis had any impact in the prognosis of CRC patients. We detected LIF expression in all samples from Marisa (Marisa et al, 2013) and TCGA (The TCGA Portal) datasets, which were significantly higher in tumors compared with normal adjacent mucosa (Figs 7G and EV5J). Gene Set Enrichment Analysis (GSEA) uncovered NF‐κB as the main activated pathway in high‐LIF tumors followed by interferon response and the JAK/STAT pathway (Fig 7H–J).

LIF high tumors were primarily ascribed to the CMS1 subtype, in stage II CRC patients, according to the Guinney classification (Guinney et al, 2015; Fig 7K) that comprises the majority microsatellite instability (MSI) tumors, which are characterized by activation of immune evasion pathways (Llosa et al, 2015). These results position LIF high tumors as candidates for immunotherapy treatment likely in combination with LIF or STAT3 inhibitors.

Further bioinformatic analysis demonstrated that LIF levels were sufficient to stratify patients displaying the poorest disease‐free survival both at all stages (P = 0.017; HR = 1.79; Fig 7L) and in the subset of stage II patients (P = 0.0027; HR = 3.08; Fig 7M), which is particularly relevant in the clinical context. Then, we studied the possible correlation between CHUK (the gene codifying for IKK‐α) and LIF levels in CRC. We observed that LIF high and low tumors were similarly distributed in the CHUK high and low groups (Fig EV5K) with no significant correlation between CHUK and LIF mRNA levels (Fig EV5L and M), suggesting that LIF expression can be regulated independently of IKKα in human tumors. Importantly, the prognosis value of LIF was restricted to tumors carrying high CHUK expression (Fig 7N and O), being patients with CHUK and LIF high tumors the ones with the worst disease‐free survival (HR = 8.31, P = 0.016; Fig 7O). These results suggested that pro‐tumorigenic activity of LIF is sustained by additional IKK‐α activities such as BRD4 or ATM regulation and postulate LIF and IKK‐α/NF‐κB as candidate prescription biomarkers for immunotherapy.

Together, our data support a model (see Fig 8) in which IKK‐α imposes therapeutic refraction in CRC cells at different levels. On the one hand, IKK‐α kinase activity facilitates DNA repair downstream of ATM (Colomer et al, 2019) and BRD4 (Li et al, 2018). On the other hand, IKK‐α through NF‐κB promotes LIF transcription and JAK/STAT pathway activation to prevent early apoptosis upon DNA damage and to facilitate STAT3/BRD4 association. Because high‐LIF levels are largely detected in the subset of CMS1 (MSI) tumors and its prognosis value is linked to the presence of high CHUK mRNA, we propose CHUK/IKK‐α and LIF as potential biomarkers of immune evasion in MSI CRC, which should be further investigated in the context of immunotherapy.

Figure 8. Model for IKK‐α‐mediated therapy resistance promotion.

In response to DNA damaging treatment, few cancer cells enter apoptosis (brown cells), which is partially prevented by STAT3 activation downstream of IKK‐α, whereas other cells accumulate variable amounts of DNA damage that is counteracted by activation of IKK‐α‐ and ATM‐dependent DDR pathway. Inhibition of STAT3 by ruxolitinib or anti‐LIF antibody and the DDR pathway by BRAF inhibitor or JQ1 (or abrogation of both pathways by IKK‐α depletion) results in massive cell death after CT treatment (dead cells are marked in dark gray).

Discussion

We had previously demonstrated that IKK‐α is an upstream regulator of the DDR pathway and cells carrying active IKK display higher resistance to CT agents linked to a more efficient DNA repair (Colomer et al, 2019). We now show that IKKα also provides CT refraction by inhibiting apoptosis, previous to DNA repair, downstream of BRD4 and STAT3 activation. Notably, these IKK‐α functions are mainly NF‐κB independent but dependent on the kinase activity of IKK‐α on specific phosphorylation substrates. We have now shown that IKK‐α is an upstream kinase of the BET protein BRD4 at S1117, which regulates BRD4 capacity for dissociation and reassociation to chromatin. This result is specifically relevant since S1117 is located at C‐terminal domain (CTD) of BRD4, which functions as a transcriptional coactivator domain by binding the transcription elongation factor (p‐TEFb) complex that regulates RNA polymerase (Pol II) activity and controls productive transcription elongation (Itzen et al, 2014). Thus, the different cycling dynamics of BRD4 observed across cell lines and damaging agents may reflect differences in the levels and dynamics of transcriptional coregulators such as transcription factors, activated STAT3 or polymerase II, or in the patterns of histone H3 and H4 acetylation. Moreover, comparison of the BRD4 sequences across species demonstrated a complete conservation of S1117 residue thus reinforcing the idea that S1117 phosphorylation may impact on BRD4 activity. This possibility opens a new avenue of research focused on the cooperative contribution of BRD4 and IKK‐α to the DNA damage response, with putative implications in cancer therapy and in the design of novel treatment protocols. To our view, the broad spectrum of IKK‐α activities, other than NF‐κB activation, points out this kinase as a preferential target for therapy in combination with DNA damaging agents as we previously proposed (Colomer et al, 2019). Using the intestinal adenoma model of APC^Min/+ murine organoids, we uncovered a novel association between BRD4 and STAT3 at chromatin. Although this is not the best model for studying human CRC (more similar to the human familial adenomatous polyposis adenoma model), the functional link between BRD4 and STAT3 downstream of IKK‐α has been further validated in several cellular systems in this work. Moreover, because proinflammatory cytokines are canonical upstream activators of IKK, we anticipate that the mechanism of therapeutic refraction imposed by the IKK‐α/JAK/STAT/BRD4 and the IKK‐α/ATM axes could be specifically relevant in tumors with high contribution of the immune system (such as tumors with high microsatellite instability, MSI), and IKK‐α‐based therapy could be even enhanced following immunotherapy treatment (i.e. inhibitors of PD1/PDL1). In this sense, we demonstrated that conditioned media from IKKα WT MEFs imposes a robust activation of STAT3 in IKK‐α‐deficient cells, suggesting that factors produced by the stroma could be inducing resistance pathways in the adjacent tumor cells. Thus, we propose that targeting IKK or JAK/STAT3 will be particularly useful for potentiating the effect of immunotherapy protocols such as combinations of chemotherapy plus PD1 or PDL1 inhibitors that are currently being used for treating MSI high CRC tumors (Oliveira et al, 2019).

Physical association between IKK‐α and STAT3 was previously found to impose IKK‐α stabilization and NF‐kB activation by preventing its association to ubiquitin ligases (Hahn et al, 2020). Now, we identified BRD4 and STAT3 as direct downstream effectors of IKK, which represents additional evidence of its prominent role in cancer, being BRD4 and STAT3 well‐known tumor and therapy resistance drivers. Importantly, STAT3 activation by IKK‐α is imposed by paracrine activation of the upstream JAK kinases by LIF. Thus, it is likely that other important factors downstream of LIF such as STAT1 or elements of the MAPK or PI3K pathway (Christianson et al, 2021) could also contribute to the observed phenotype. Because LIF has recently been identified as a suppressor of anti‐PD1 therapy (Pascual‐García et al, 2019; Hallett et al, 2023) and we found a clear association between LIF levels in CRC with MSI tumors and the immune evasion phenotype, we propose that inhibitors of IKK activity could have particular impact in tumors that are candidate for immunotherapy treatment. In the absence of safety and highly specific IKK‐α inhibitors that have already been approved for human therapy (reviewed in Colomer et al, 2020), we propose that combination treatments involving CT plus BRAF inhibitors and/or inhibitors of the JAK/STAT and BRD4 pathways will represent a suitable strategy to revert chemo‐refraction, thus leading to more efficient therapeutic protocols.

Materials and Methods

Patient‐derived and mouse intestinal organoids

For patient‐derived organoids (PDOs) generation, primary or xenografted human colorectal tumors were disaggregated in 1 mg/ml collagenase II (Sigma) and 20 μg/ml hyaluronidase (Sigma), filtered in 100 μm cell strainer, and seeded in Matrigel (BD Biosciences) as described (Sato et al, 2011). PDOs were expanded by serial passaging and kept frozen in liquid Nitrogen for being used in subsequent experiments. PDO5 is TP53 WT and carries the KRAS G12D mutation. PDO8 is TP53 Q192stop and KRASG13C. APC^Min/+‐derived organoids from WT or IKK‐α KO mouse intestines were obtained as previously described (Colomer et al, 2018).

Ethics

Samples from patients were kindly provided by MARBiobank, integrated in the Spanish Hospital Biobanks Network (RetBioH; www.redbiobancos.es). Informed consent was obtained from all participants and protocols were approved by institutional ethical committees (Ethical Committee approval #2019/8595/I).

Cell lines

CRC cell lines HCT116 (KRAS mutated), Caco2 (KRAS and BRAF WT) and HT29 (BRAF mutated) were obtained from the American Type Culture Collection (ATCC, USA). WT and IKK‐α KO MEFs were kindly provided by Michael Karin (UCSD, La Jolla), and Hela S3 cells were kindly provided by Dr. Shannon Lauberth, UCSD. All cells were grown in Dulbecco's modified Eagle's medium (Invitrogen) plus 10% fetal bovine serum (Biological Industries) and were maintained in a 5% CO₂ incubator at 37°C. Cells were routinely tested by PCR as being mycoplasma free.

IKK1 CRISPR knockout cells were generated by CRISPR‐Cas9. Guides were designed using the online prediction tool created by Zhang lab at Massachusetts Institute of Technology (http://www.e‐crisp.org/E‐CRISP/designcrispr.html). Two guides located between Exon 1 and Intron 1 on human CHUK gene (IKK1) with few predicted off targets were cloned into the LentiCRISPR V2 plasmid (Addgene plasmid#52961).

Animal studies

To perform in vivo drug testing, equivalent pieces of individual tumors were implanted orthotopically in the wall of the cecum of athymic nude mice. When tumors were detectable by palpation (4–5 weeks), animals were randomly ascribed to the different groups of treatment. Vemurafenib (50 mg/kg) and ruxolitinib (100 mg/kg) were administered orally every day, 5‐FU (75 mg/kg, divided in two doses) and irinotecan (20 mg/kg) every 4 days intravenously. After 21 days of treatment or at the humanitarian endpoint (in the case of the long‐term survival experiment), mice were euthanized and tumors collected, photographed, measured, and processed for immunohistochemistry examination. In all our procedures, animals were kept under pathogen‐free conditions, and animal work was conducted according to the guidelines from the Animal Care Committee at the Generalitat de Catalunya. The Committee for Animal Experimentation at the Institute of Biomedical Research of Bellvitge (Barcelona) approved these studies.

PDOs infection

sgRNA against CHUK gene was designed using Benchling. Lentiviral production was performed transfecting in HEK293T cells the lentiviral vectors and the plasmid of interest. One day after transfection, medium was changed, and viral particles were collected 24 h later and then concentrated using Lenti‐X Concentrator. PDOs were infected by resuspending single cells in concentrated virus diluted in complete medium, centrifuged for 1 h at 650 rcf, and incubated for 5 h at 37°C. Cells were then washed in complete culture medium and seeded as described above.

PDO viability assays

Six hundred single PDO cells were plated in 96‐well plates in Matrigel. After 6 days in culture, we treated growing PDOs with 5‐FU, irinotecan, AZ628, ruxolitinib or combinations for 72 h at the indicated concentrations. Cell viability was determined using the CellTiter‐Glo 3D Cell Viability Assay (Promega) following manufacturer's instructions in an Orion II multiplate luminometer (Berthold detection systems). Data were calculated as mean ± standard deviation, representing triplicates of one out of two independent experiments.

Cell lysis and Western blot (WB)

Cells were lysed 20 min at 4°C in 300 μl of PBS plus 0.5% Triton X‐100, 1 mM EDTA, 100 mM NA‐orthovanadate, 0.25 mM phenylmethylsulfonyl fluoride, and complete protease inhibitor cocktail (Roche). Lysates were analyzed by Western blotting using standard SDS–polyacrylamide gel electrophoresis (SDS–PAGE) techniques. In brief, protein samples were boiled in Laemmli buffer, run in polyacrylamide gels, and transferred onto polyvinylidene difluoride membranes. The membranes were incubated overnight at 4°C with the appropriate primary antibodies. After being washed, the membranes were incubated with specific secondary horseradish peroxidase‐linked antibodies from Dako and visualized using the enhanced chemiluminescence reagent from Amersham. Primary antibodies used are listed in Table EV1.

Cell fractionation

For cytoplasm/nuclear/chromatin separations, cells were lysed in 10 mM Hepes, 1.5 mM MgCl₂, 10 mM KCl, and 0.05% NP‐40 (pH 7.9) for 10 min on ice and centrifuged at 900 g. Supernatants were recovered as the cytoplasmic fraction, and the pellets were lysed in 5 mM Hepes, 1.5 mM MgCl₂, 0.2 mM EDTA, 0.5 mM dithiothreitol, and 26% glycerol and sonicated for 5 min three times to recover the soluble nuclear fractions. The remaining pellet included the chromatin fraction. Lysates were run in SDS–PAGE and transferred onto Immobilon‐P transfer membranes (Millipore) for Western blot analysis.

Immunoprecipitation assay (IP) and pull‐down assay (PD)

PD assays were performed as previously described (Espinosa et al, 2003a). Briefly, GST fusion proteins were incubated with lysates for 45 min in a rotary shaker at 4°C. When indicated, nuclear extracts were boiled at 98°C for 5 min in the presence of 1% SDS to disassemble pre‐existing protein complexes and then neutralized in 1% Triton X‐100. Precipitates were resolved in SDS–PAGE and analyzed by IB. For peptide IP, histone H4 peptides [Synpeptide CO LTD] were synthesized as biotinylated N‐terminal and C‐terminal amides. Peptides were incubated overnight at 4°C with the indicated cell extracts and precipitated with streptavidin–sepharose beads for 45 min.

Mass spectrometry analysis

Cell lysates obtained in the different experimental conditions were processed and digested with trypsin and endoproteinase LysC with a ratio enzyme:sample of 1:10 for both enzymes (w:w). Samples were then subjected to phosphopeptide enrichment using titanium dioxide (TiO₂) beads, and phospho‐enriched samples were analyzed by LC–MS/MS. To identify IKKα‐dependent phosphopeptides, samples were injected with a 120‐min chromatographic gradient in an Orbitrap Velos Pro with a data‐dependent acquisition method using CID fragmentation for the top 20 most intense precursor ions and multistage activation. In the UV‐activation experiment, samples were acquired with a 90‐min gradient in an Orbitrap Fusion Lumos with a data‐dependent acquisition method using top speed, HCD fragmentation, and ion‐trap detection. In both cases, the resulting data were analyzed with the Proteome Discoverer software v1.4, using the search algorithm Mascot (v2.5) against a Human protein database (Uniprot, v2015) with oxidation (Met), and phosphorylation (Ser, Thr, Tyr) as variable modifications. Carbamidomethylation (Cys) was set as fixed modification and a mass tolerance of 7 ppm (MS1) and 0.5 Da (MS2) were used. Only peptides with a false discovery rate below 5% were considered for quantitative analysis. Peptides' relative abundance was estimated with the area under the curve of extracted ion chromatograms. Protein network was generated using cytoscape software (www.cytoscape.org).

ChIP sequencing and data analysis

DNA samples were sequenced using Illumina HiSeq platform. Raw single‐end 50‐bp sequences were filtered by quality (Q > 30) and length (length > 20 bp) with Trim Galore (Krueger, 2012). Total filtered sequences, which ranged between 32 and 55 million per sample, were aligned against the reference genome (mm10 release) with Bowtie2 (Langmead & Salzberg, 2012). MACS2 software (Zhang et al, 2008) was run for each replicate considering unique alignments (q‐value < 0.1). Peaks from biological replicates were merged using bedtools. Peaks that were detected in the input samples were filtered out, as well as the mouse black regions downloaded from the ENCODE portal (ENCFF547MET; Sloan et al, 2016; https://www.encodeproject.org/). Peak annotation was performed with ChIPseeker (Yu et al, 2015) and Annotatr (Cavalcante & Sartor, 2017) packages; and functional enrichment analysis with enrichR (Kuleshov et al, 2016), using the latest version of GO annotations. Peaks heatmaps and multiple coverage correlation across bigwig files were performed with Deeptools (Ramírez et al, 2016).

Bioinformatics analysis

The transcriptome of normal and tumor samples from GSE39583 (Marisa et al, 2013) obtained by microarrays was downloaded from the Gene Expression Omnibus (GEO; Edgar et al, 2002) and analyzed with the affy R package. Transcriptomic and available clinical data from CRC TCGA dataset were downloaded from the open‐access resource CANCERTOOL. We used the TCGA cohort (The TCGA Portal) to classified patients according to the mean expression of LIF. The association with relapse was assessed using Kaplan–Meier estimates and Cox proportional hazard models. A standard log‐rank test was applied to assess significance between groups. This test was selected because it assumes the randomness of the possible censorship. All the survival analyses and graphs were performed with R using the survival (v.3.2‐3) and survimer (v.0.4.8) packages, and a P‐value < 0.05 was considered statistically significant.

Differentially expressed genes between patients with high versus down levels of LIF were explore using DESeq2 R package (v.1.24.0; Love et al, 2014) from gene expression data downloaded from TCGAbiolinks R package (Colaprico et al, 2016). Functional enrichment analysis was performed with enrichR, using Gene Ontology biological process annotation.

Plasmids

Plasmid plentiCRISPRv2 was a gift from Feng Zhang (Addgene plasmid #52961; http://n2t.net/addgene:52961; RRID:Addgene 52961); plasmid p6344 pcDNA4‐TO‐Ha‐Brd4FL, a gift from Peter Howley (Addgene plasmid #31351; http://n2t.net/addgene:31351; RRID:Addgene_31351); High fidelity Prime STAR HS DNA polymerase was from TAKARA. Oligonucleotides were purchased from sigma. The sequence of the different oligonucleotides used to generate the different constructs is listed in Table EV2.

Plasmid construction

GST fusion proteins were generated using the vector pGEX5x.3. BRD4 and STAT3 coding sequences were cloned in frame downstream from GST, using BamHI and XhoI restriction sites. Sequences of interest were amplified by PCR using appropriate primers (Table EV2). Primers contain 5′ extensions to include restriction sites and to ensure the correct reading frame.

For STAT3, two fragments were fused to GST, amino acids 507–700 and 559–700. The STAT3 coding sequence was taken from Genebank, accession number NM_001369512. The starting material for cloning was cDNA prepared from HT29 cells RNA.

Several BRD4 fragments were produced as GST fusion proteins in E. coli. These include amino acids 2–320; 301–700; 685–1,000, 1,001–1,362, 526–702, 577–641, and 1,037–1,201. To generate the appropriate plasmids, PCR was used on the template plasmid p6344 pcDNA4‐TO‐Ha‐Brd4FL. The mutagenic primers used to generate the point mutation in codon 1,117, that resulted in the substitution of serine with alanine, are depicted in Table EV2. Mutagenesis was performed by sequence overlap extension. For the full‐length coding sequence, a fragment replacement was done taking advantage of a unique AgeI restriction site. All DNA constructs were verified by Sanger sequencing.

To modify the BRD4 gene in situ, two guide RNAs were designed using the Benchling platform (https://www.benchling.com/crispr/). One guide targets the double DNA cut 27 nucleotides 5′ of codon 1,117 and the other 5 nucleotides 3′ of the stop codon. The information to produce these guide RNAs (primers gU and gD1117 in Table EV2) was introduced in plasmids pLentiCRISPR V2‐BRD4.1 and pLentiCRISPR V2‐BRD4.2 following published procedures (Sanjana et al, 2014). In addition, a repair plasmid, pBS‐BRD4S1117A‐ki, was constructed to achieve homologous recombination next to the targeted chromosomal locations. This plasmid carries two homology arms which extend 847 bp upstream and 906 bp downstream of the fragment to be replaced. The replacement DNA contains the desired codon 1,117 mutation and several silent changes that rend the edited chromosome‐resistant non‐recognizable by the guide RNAs. Moreover, this DNA fragment contains intronic and exonic sequences up to the BRD4 stop codon, which is preceded by a loxP‐P2A‐mEmerald‐loxP cassette. The mEmerald coding sequence was amplified from pmEmerald‐Claudin7‐C‐12.

CRISPR/Cas9‐mediated BRD4 gene editing

HCT116 cells were co‐transfected with the described targeting plasmids and the repair plasmid at a 1:1:40 molecular ratio. After 24‐h incubation, puromycin (1 μg/ml) was added for 72 h to enrich for transfected cells and then removed. Cells were expanded for about 2 weeks and sorted and plated as single cells in 96‐well plates to isolate clones.

Kinase assay

In vitro kinase assays were performed as previously described (Espinosa et al, 2003b). In brief, 5 μg of GST or GST fusion proteins was incubated with 200 ng of recombinant human IKKα (ab102103) or 10 μg of cell lysates, as indicated, at 30°C for 30 min in the presence of ATP‐γP³². Reactions were stopped by adding loading buffer, run in a polyacrylamide gel, and developed in autoradiograph film. Primers used to generate GST fusion proteins are listed in Table EV2.

RT–qPCR analysis

Total RNA from treated APC^Min/+‐derived organoids and MEFs cells was extracted with the RNeasy Micro Kit, and cDNA was produced with the RT‐First Strand cDNA Synthesis Kit. RT–qPCR was performed in LightCycler 480 system using SYBR Green I Master Kit. Samples were normalized to the mean of the housekeeping genes GAPDH and ACTB. Primers used for RT–qPCR are listed in Table EV3.

Hematoxylin and eosin staining

Previously de‐paraffinized sections were incubated with hematoxylin 30 s, tap water 5 min, 80% ethanol 0.15% HCl 30 s, water 30 s, 30% ammonia water (NH3(aq)) 30 s, water 30 s, 96% ethanol 5 min, eosin 3 s, and absolute ethanol 1 min. Samples were dehydrated, mounted in DPX, and images were obtained with an Olympus BX61 microscope.

Immunohistochemical staining

Paraffin blocks were obtained from tumor samples, previous fixation in 4% formaldehyde overnight at room temperature. Paraffin‐embedded sections were de‐paraffinized and rehydrated, and endogenous peroxidase activity was quenched (20 min, 1.5% H₂O₂). Citrate‐based antigen retrieval was used. All primary antibodies were diluted in PBS containing 0.05% BSA, incubated overnight at 4°C, and developed with the Envision⁺ System HRP Labeled Polymer anti‐Rabbit or anti‐Mouse and 3,3′‐diaminobenzidine (DAB). Samples were mounted in DPX, and images were obtained with an Olympus BX61 microscope.

Statistical analysis

Statistical parameters, including number of events quantified, standard deviation, and statistical significance, are reported in the figures and in the figure legends. Statistical analysis has been performed using GraphPad Prism6 software (GraphPad), and P < 0.05 is considered significant. Two‐sided Student's t‐test was used to compare differences between two groups and two‐way ANOVA test was used to compare differences among multiple groups. Each experiment has been repeated at least twice.

Author contributions

Lluís Espinosa: Conceptualization; data curation; supervision; funding acquisition; validation; investigation; methodology; writing – original draft; writing – review and editing. Irene Pecharromán: Formal analysis; validation; investigation; visualization; methodology. Laura Solé: Supervision; investigation; methodology; writing – original draft; writing – review and editing. Daniel Álvarez‐Villanueva: Investigation; methodology. Teresa Lobo‐Jarne: Data curation; software; funding acquisition; investigation; methodology. Josune Alonso‐Marañón: Investigation; methodology. Joan Bertran: Conceptualization; investigation; methodology; writing – review and editing. Eduard Sabidó: Supervision; investigation; methodology; writing – review and editing. Eva Borràs: Supervision; investigation; methodology; writing – review and editing. Violeta García‐Hernández: Investigation; methodology; writing – review and editing. Marta Garrido: Investigation; methodology. Joan Seoane: Supervision; validation; writing – review and editing. Anna Bigas: Conceptualization; resources; supervision; writing – original draft; writing – review and editing. Alberto Villanueva: Supervision; investigation; methodology; writing – review and editing. María Martínez‐Iniesta: Investigation. Raffaella Iurlaro: Investigation; methodology. Cristina Santos: Investigation; methodology. Yolanda Guillén: Data curation; software; formal analysis; validation. Ángela Montoto: Investigation; methodology. Gemma Giménez: Investigation; methodology. Mar Iglesias: Formal analysis; investigation; writing – review and editing. Ramon Salazar: Supervision. Ester Bonfill‐Teixidor: Investigation; methodology.

Disclosure statement and competing interests

The authors declare that they have no conflict of interest.

Supporting information

Expanded View Figures PDF

Table EV1

Table EV2

Table EV3

Dataset EV1

PDF+

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 3

Source Data for Figure 4

Source Data for Figure 5

Source Data for Figure 6

Source Data for Figure 7

The EMBO Journal (2023) 42: e114719

Data availability

MS data are available at PRIDE EBI‐EMBL database with identifier PXD008932 and in Colomer et al (2019).

ChIP‐seq data for BRD4 are deposited at GEO (GSE196461; http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE196461).

Microscopic images from Fig 6 are deposited at BioStudies (S‐BIAD844; https://www.ebi.ac.uk/biostudies/studies/S‐BIAD844).