CDX2 inducible microRNAs sustain colon cancer by targeting multiple DNA damage response pathway factors Free

MicroRNAs (miRs) are among the many important factors that control gene expression. The role of miRs in genome instability and ultimate progression to different types of cancers has been well studied. It is interesting to note that changes in the miR expression levels occur during the above processes by generating a threshold in target gene expression (Mukherji et al., 2011). Microarray expression or small RNA sequencing data from a large number of different cancers show both increased or decreased miR levels (Chung et al., 2017; Croce, 2009), which results in transcriptional rewiring of the cognate genes. A significant number of miRs have been reported to be dysregulated in multiple types of cancers, including colorectal carcinoma (CRC) (Chen et al., 2019; Ding et al., 2018).

To prevent the development of cancers, maintenance of genomic integrity is of utmost importance. Normal cells initiate the DNA damage response (DDR) pathway when exposed to multiple types of DNA damages, such as stalled replication, ionizing irradiation (IR), ultraviolet lights and genotoxic drugs (Ciccia and Elledge, 2010; Tikoo and Sengupta, 2010). One of the key factors that regulates genome integrity is BLM. BLM mutations lead to Bloom syndrome (BS) and are associated with DNA damage sensing, DNA repair, recombination (Sengupta et al., 2004; Tikoo et al., 2013; Tripathi et al., 2018) and cancer predisposition (Cunniff et al., 2017; Kaur et al., 2021). MRE11-RAD50-NBN (the MRN complex) and ATM are upstream DDR factors that recognize and accumulate at the double-strand breaks within seconds after IR exposure (Smith et al., 2010; Syed and Tainer, 2018). Multiple other factors take part in this choreographed process, including E3 ligases, such as RNF8 (which facilitate recruitment of DDR proteins) (Zhou et al., 2019), damage-specific kinases, such as Chk1 and Chk2 (also known as CHEK1 and CHEK2, respectively; Smith et al., 2010), and BRCA1. Tumor suppressor BRCA1 maintains genome stability by being a part of multiple protein complex and is involved in DNA damage repair, DNA damage-induced cell cycle checkpoint activation, protein ubiquitylation, transcriptional regulation and apoptosis (Savage and Harkin, 2015). Loss of any of the genome stabilizers (such as BLM, ATM and BRCA1) leads to accumulation of DNA damage, which progresses to neoplastic transformation, cancer predisposition and finally the development of cancer (Hanahan and Weinberg, 2011; Negrini et al., 2010). Links between miRs and BRCA1 have been documented previously (Chang and Sharan, 2012; Petrovic et al., 2017). For example, miR-182 has been shown to downregulate BRCA1, impacting DNA repair and sensitivity to PARP inhibitors (Moskwa et al., 2011).

In this context, we performed meta-analysis of the The Cancer Genome Atlas (TCGA) database and identified DNA damage sensitive miRs (DDSMs), which are upregulated across all stages of colon adenocarcinoma tissue samples. We validated DDSMs in an independent cohort of colon adenocarcinoma patient tissues and in multiple cellular models with high levels of endogenous DNA damage. All of the DDSMs were regulated by a common transcription factor elevated in the colon, CDX2. In normal cells, transcription of CDX2 was repressed due to BLM-dependent recruitment of HDAC1/2-containing Sin3b and NuRD complexes on its promoter. DDSMs target multiple key proteins involved in DDR (BRCA1, ATM, Chk1 and RNF8). Specifically, DDSMs downregulate BRCA1 expression in cancer cell lines and colon adenocarcinoma patient tissue samples, thereby allowing neoplastic transformation. Hence, this report serves as an integrated study in which we have demonstrated how a patient-derived colon cancer-specific miR signature is epigenetically silenced in normal tissues and becomes deregulated during DNA damage, thereby repressing DNA damage response factors and subsequently worsening the pathological consequence.

To identify the miRs that were upregulated in all stages of colon adenocarcinoma, we carried out an unbiased meta-analysis of the TCGA database. We found that of the total 1323 miRs for whom the transcript expression values were available, only 647 miRs were found to be expressed in at least 50% of samples. Differential expression analysis indicated that 382 miRs were downregulated and 138 miRs were upregulated compared to the normal samples. Of the 138 miRs that were upregulated, the transcript levels of 62 miRs were increased all across the four different stages of colon cancer, which was depicted in the form of a heatmap (Fig. S1). All of the predicted targets for these 62 miRs were then analyzed by miRanda, miRTarBase and Target Scan. The common targets were then analyzed by four pathway analysis software packages [Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), Wiki Pathways and Gene set Enrichment Analysis (GSEA)]. Ten upregulated miRs were enriched in pathways that were associated with colorectal cancer and were directly implicated in DDR, namely DNA repair, apoptosis, DNA replication and chromatin remodelling. Recognizing the fact that intrinsic DNA damage is enhanced within the cancerous lesions, we named these miRs as DDSMs (Fig. 1A). We randomly chose six of the DDSMs (miR-29a-5p, miR-29b-3p, miR-96-5p, miR-182-5p, miR-183-5p and miR-335-3p) for further studies. Among the 62 miRs that were upregulated across four stages of cancers, we also chose three miRs (miR-130a-5p, miR-148a-5p and miR-429), which did not fit the above criteria for DDSMs. These three miRs were called non-DDSMs.

The six chosen DDSMs were significantly upregulated in the tissues of the colon cancer patients in the TCGA database across all the four stages of cancer progression (Fig. 1B). Next, we expanded our studies to 54 paired tissue samples obtained from an Indian cohort. These six DDSMs were upregulated in the patient cancerous tissue samples compared to their matched adjacent normal tissues across various stages of cancer progression (Fig. 1C). Kaplan–Meier analyses indicated lesser overall survival for patients with a higher risk score (i.e. patients that show higher levels of the combined expression of the six miRs) in the tissue samples (Fig. 1D).

To further investigate the functions of DDSMs, we chose cells from BS patients that show high levels of endogenous DNA damage (Kharat et al., 2016), hyper-recombination, genome instability and increased sister chromatid exchanges (SCEs) (Cunniff et al., 2017). BS patients are predisposed to multiple types of cancers. The most common type of solid tumour in BS patients is colorectal cancer (Cunniff et al., 2017). Further, when BLM is heterozygous, there is an enhanced risk of colon cancer in both mice and human (Goss et al., 2002; Gruber et al., 2002). Hence, we carried out small RNA sequencing with RNA isolated from two pairs of isogenic cell lines created from two different BS patients (see Materials and Methods for details about isogenic lines). Small RNA sequencing indicated that the DDSMs were upregulated in cells lacking BLM (Fig. 2A; Fig. S2A). The higher expression of the six identified DDSMs was validated by RT-qPCR in BLM-deficient cells compared to BLM-restored cells in both the above isogenic pairs (Fig. 2B; Fig. S2B). We further demonstrated the increased expression of the DDSMs in BLM knockout HCT116 cells compared to parental cells (Fig. 2C), as well as in siRNA-mediated BLM knockdown SW480 cells compared to control cells (Fig. S2C).

Next, we wanted to determine whether the matured DDSMs were also enriched in the RNA-induced silencing complex (RISC) under the same conditions. HeLa S3 cells stably integrated with either vector pREV or Flag Ago2 were used for the experiment. BLM was ablated in both the cell lines using BLM siRNA (Fig. 2D). In either the absence or presence of BLM, immunoprecipitations were carried out with either anti-GFP (which acted as an antibody control) or with anti-Flag antibody. RNA bound to the immunoprecipitated Ago2 complex was isolated followed by RT-qPCR, which showed that the five tested DDSMs were associated with Ago2 protein (i.e. the RISC complex) (Fig. 2E).

Absence of BLM leads to persistence of damaged DNA, which culminates in hyper-recombination (Croteau et al., 2014; Tikoo and Sengupta, 2010). Hence, we hypothesized that in a self-perpetuating loop, the DDSMs might respond to the increased levels of cellular DNA damage. To test this hypothesis, BLM isogenic cells were exposed to a range of IR doses (leading to increased DSB generation). Although the levels of the DDSMs increased over the range of IR dosage (Fig. 2F), the three non-DDSMs did not respond to the IR gradient (Fig. S2D).

As the levels of DDSMs depend on the extent of DNA damage, we next wanted to determine whether the DDSMs can potentiate neoplastic transformation in both in vitro and in vivo models. We found that the ablation of DDSMs by miR inhibitors in the BLM-deficient GM03509 GFP100 cells caused a decrease in the extent of residual DNA damage, as determined by neutral comet assays (Fig. 3A) and spontaneous levels of SCEs (Fig. 3B). Reciprocally, overexpression of the same DDSMs in the BLM-restored GM03509 GFP-BLM Clone 4.3.4 cells led to an enhancement in SCEs (Fig. 3C). Subsequently, inhibition of two DDSMs (miR-29a-5p and miR-96-5p) in HCT116 BLM knockout cells culminated in a decrease in the invasive potential of these cells, as determined using both matrigel (Fig. 3D) and soft agar (Fig. 3E) assays.

To determine whether modulation of the levels of DDSMs affected the ability to initiate and propagate tumors, we generated three stable lines in HCT116 BLM^−/− cells, which expressed either a control inhibitor or specific inhibitors of miR-29a-5p or miR-96-5p. Each of the stable lines, which also constitutively expressed EGFP (Fig. S3A), were validated for lowering levels of miR-29a-5p or miR-96-5p (Fig. S3B). Xenograft assays were carried out in NOD SCID mice by subcutaneously injecting these cells in the animals and monitoring for tumor development. Results indicated that inhibition of the two DDSMs decreased the rate of tumor growth (Fig. 3F). As a second in vivo assay, 100 mm³ tumors were subcutaneously generated using HCT116 BLM^−/− cells. A nanoparticle-mediated delivery system was used to deliver miR inhibitor control, miR inhibitor 29a-5p or miR inhibitor 96-5p directly to the base of the tumors every third day. Tumor formation was decreased when miR inhibitor 29a-5p or miR inhibitor 96-5p were injected (Fig. 3I). Analysis of the levels of miR-29a-5p and miR-96-5p in these tumors (excised at the endpoint in both the experiments) confirmed the decrease in the levels of the two tested DDSMs (Fig. 3G,J). Thus, the DDSMs are miRs that sustain tumor growth.

Next, we wanted to determine how the DDSMs were regulated in the cellular milieu. Using an in silico approach, we first analyzed the upstream 5 kb promoter of each of these miRs. We found that only one transcription factor, CDX2, was common among the DDSMs (Table S1). We hypothesized that CDX2 (known to be highly expressed in colon) may control the expression of the DDSMs. We observed CDX2 expression to be slightly increased 1 h after 3 Gy IR exposure in HCT116 cells. This effect was significantly enhanced by the loss of BLM in HCT116 BLM knockout cells. However, 6 h after IR exposure (when a certain amount of DNA repair has occurred) the levels of CDX2 began to decrease (Fig. 4A). An electrophoretic mobility shift assay (EMSA) demonstrated that CDX2 bound to the promoter sequence of the tested DDSM, miR-96-182-183. The binding was specific as a supershift was observed when an anti-CDX2 antibody was used in the EMSA reactions (Fig. 4B). The binding of CDX2 to the miR promoter was lost in the presence of an excess of unlabeled competitor (Fig. 4C) or when a mutant oligo in which the CDX2 binding site was destroyed was used as the substrate (Fig. S3D). Using immunoprecipitated (IPed) CDX2, it was shown that the binding of CDX2 to the miR promoter was enhanced 1 h post-IR (Fig. 4D). The ability of CDX2 to transactivate the DDSMs in a dose-dependent manner was demonstrated for the full-length protein but not for the mini CDX2 splicing variant lacking the transactivation domain (Fig. 4E). Mutating the CDX2 binding site on the miR-96-182-183 promoter prevented its transactivation by CDX2 (Fig. 4F). Finally, ablation of CDX2 by siRNA (Fig. 4G) decreased the levels of primary, precursor and mature miRs within the cells (Fig. 4H; Fig. S3E).

We next hypothesized that modulating the levels of CDX2 should also lead to changes in the DDSM levels and thereby the control in the tumor formation process. First, we used HT29-derived colon cancer cells with doxycycline (Dox) inducible expression of CDX2 plus GFP (TW6 cell line) or only GFP (TG8 cell line) (Fig. 5A). Induction of CDX2 by Dox treatment in TW6 cells (Fig. 5A) led to an increase in the levels of all the six DDSMs (Fig. 5B), which was concomitant with enhanced wound healing (Fig. 5C) and colony formation (Fig. 5D) efficiencies. The stimulatory effect of CDX2 was also tested in another pair of cell lines based on HCT116 cells – namely HW2 (constitutively expressing Flag CDX2) and HC1 (the corresponding vector control) (Fig. S4A). Overexpression of Flag CDX2 in HW2 cells led to the upregulation of the DDSMs (Fig. S4B). Next, in vivo experiments were carried out using xenograft models in which TG8 and TW6 cells were subcutaneously injected into NOD SCID mice, a subset of which were fed orally with Dox. Tumors obtained from TW6 cells expressing inducible CDX2 had the maximum tumor volume (Fig. 5E). Tumors were excised at the endpoint of the above experiment. Tumors obtained from TW6 cells showed increased levels of all the DDSMs (Fig. 5F), the proliferation marker PCNA (Fig. 5G) and angiogenesis marker CD31 (Fig. 5H). We also carried out in vivo subcutaneous xenograft studies using the HC1/HW2 cells. The sizes of tumors obtained from HW2 cells were consistently larger (Fig. S4C) and expressed higher levels of the DDSMs (Fig. S4D), PCNA (Fig. S4E) and CD31 (Fig. S4F). Finally, we wanted to determine whether expression of the DDSMs caused increased dissemination of the cancer cells. Using the TG8/TW6 lines, we generated an orthotopic model in which the cells were implanted into the cecal wall of the mice (Fig. 5I). All of the mice were orally fed with Dox. At 21 days post experiment initiation, the mice were subjected to whole-body imaging and were GFP fluorescence tracked. TW6 cells expressing CDX2 showed enhanced in vivo dissemination to distal organs (Fig. 5I).

Having established how the DDSMs are regulated in the colon, we wanted to decipher how these miRs function and sought to determine their putative targets. Using MiRanda, we found 4840 common targets for the six DDSMs (Fig. S5A, Tables S6, S7). Using Reactome, KEGG, Wiki Pathways and GSEA, we determined that the DDSMs regulated many vital cellular processes, including DNA repair, DNA replication and gene expression (Fig. S5B). From the list of common targets of the DDSMs (Table S7), we chose four known genes involved in the DDR pathways, namely BRCA1, ATM, Chk1, RNF8, for further in-depth studies. Decrease of the high levels of endogenous DDSMs by treatment with specific miR inhibitors led to an increase in the transcript levels of BRCA1 (Fig. 6A; Fig. S5C), ATM (Fig. 6B), Chk1 (Fig. 6C) and RNF8 (Fig. 6D) in GM03509 GFP Clone 100 (Fig. 6A-D) and HCT116 BLM^−/− (Fig. S5C) cells. Reciprocally, enhancement of the levels of DDSMs by treatment with specific mimics led to a decrease in the transcript levels of BRCA1 (Fig. 6E; Fig. S5D), ATM (Fig. 6F), Chk1 (Fig. 6G) and RNF8 (Fig. 6H) in both GM03509 GFP-BLM Clone 4.3.4 (Fig. 6E-H) and HCT116 wild-type (Fig. S5D) cells. The effect of the miR inhibitors and mimics at the RNA level was phenocopied at the protein levels in both the cell types tested (Fig. 6I,J; Fig. S5E,F).

We chose BRCA1 for mechanistic and clinical studies as meta-analyses have shown that mutations of this gene conferred increased risk to colon cancer (Oh et al., 2018). The RNA and the protein levels of BRCA1 were both enhanced when the miRs were downregulated in diverse experimental systems, including transient transfection of miR inhibitors in multiple cell types (Fig. 6A,I; Fig. S5C,E), stable lines expressing DDSM inhibitors (Fig. S3B,C) and in tumors obtained in two xenograft models in which expression of DDSMs are inhibited (Fig. 3H,K). Reciprocally, we also observed that the levels of BRCA1 were downregulated when the DDSM levels were enhanced in multiple experimental systems, like cells exposed to DDSM mimics (Fig. 6E,J; Fig. S5D,F), and in tumors in which the DDSM levels were increased by inducible or constitutive enhancement of CDX2 levels (Fig. 5G; Fig. S4E) and when cells were exposed to an increasing dosage of IR (Fig. S5G). Analysis of the 3′ untranslated region (UTR) of BRCA1 revealed binding sites of four of the DDSMs (Table S8). We used a luciferase reporter assay using either wild-type or mutant sequence of the BRCA1 3′ UTR, which binds to miR-183. Overexpression of miR-183-5p led to the reduction of luciferase activity when wild-type BRCA1 3′ UTR, but not mutant BRCA1 3′ UTR, was used (Fig. 6K). We hypothesized that if DDSMs act through BRCA1, increasing or decreasing BRCA1 levels should phenocopy the effects seen by modulating the levels of the miRs in cellular invasion assays (shown in Fig. 3D,E). Indeed, ablation of BRCA1 by siRNA in three different cell types [verified at both transcript and protein levels (Fig. S6A,B)] led to an increase in the levels of endogenous DNA damage (Fig. S6C), SCEs (Fig. S6D) and invasion (Fig. S6E). Conversely, overexpression of BRCA1 (Fig. S6F) decreased the DNA damage levels (Fig. S6G), SCEs (Fig. S6H) and invasion (Fig. S6I).

We next wanted to determine the regulatory circuit that controls the DDSM expression. We hypothesized that BLM itself may negatively regulate CDX2 expression as the levels of CDX2 were elevated in HCT116 BLM^−/− cells (Fig. 7A; Fig. S7A). Further, overexpression of BLM reduced the transcript level of CDX2 (Fig. 7B). We next wanted to determine whether BLM is recruited to the promoter of CDX2. Upon analysis of the 5 kb upstream of the CDX2 promoter transcription start site (TSS), we identified potential binding sites for several transcription factors – MAD, AP3β (also known as AP3B1), SMAD3, AP2β (also known as AP2B1) and E2F1 (Fig. S7B). Using ChIP, we found that BLM was specifically recruited to all these putative binding sites except to the E2F1 site (Fig. 7C). This recruitment of BLM was independent of its helicase activity as it occurred to a similar extent even in presence of ML216, a specific BLM helicase inhibitor (Nguyen et al., 2013) (Fig. S7C). Two transcriptional factors, SMAD3 and AP2β, were themselves recruited to their cognate binding sites in a BLM-dependent manner (Fig. 7D,E).

Transcriptional repression is controlled by two major protein complexes – NuRD and Sin3. These two repressor complexes have specific subunits and also share common subunits (Baymaz et al., 2015). Liquid chromatography with tandem mass spectrometry analysis of immunoprecipitated BLM from the extracts of GM03509 GFP BLM Clone 4.3.4 cells indicated its interaction with both Sin3b and CHD4, the core ATPase subunits of the two complexes. The peptide sequences found associated with BLM immunoprecipitates were SQSIDTPGVIR (for Sin3b) and APEPTPQQVAQQQ (for CHD4) (Fig. S7D,E). BLM or Sin3b immunoprecipitations further revealed that BLM interacted with Sin3b, CHD4, HDAC1 and SMAD3 (Fig. S8A-E). Direct interaction was also observed between GST-tagged BLM and in vitro transcribed and translated SMAD3, as well as GST-tagged BLM and in vitro transcribed and translated HDAC1 (Fig. S8F).

Next, we wanted to determine whether BLM is co-recruited with members of the NuRD and Sin3b complexes onto the CDX2 promoter. Using ChIP, we found that the core ATPase subunits of the two co-repressor complexes, Sin3b and CHD4, are recruited to different binding sites on the CDX2 promoter. Although Sin3b was recruited to the two SMAD3 binding sites (Fig. 7F), CHD4 was recruited to one of the two AP2β sites (Fig. 7G). The extent of recruitment of both Sin3b and CHD4 was always enhanced in cells that expressed BLM. HDAC1 and HDAC2 are the two common factors present in both NuRD and Sin3b complexes. We found that although HDAC1 was recruited exclusively to SMAD3 binding sites (Fig. 7H), HDAC2 was recruited to the regions in which MAD, AP3β, SMAD3 and AP2β binding sites were present (Fig. 7I) in a BLM-dependent manner. Sequential re-ChIP experiments validated that BLM-Sin3b and BLM-CHD4 were binding to specific DNA recognition sequences on the CDX2 promoter (Fig. S8G,H). Based on these results, we hypothesized that BLM repressed CDX2 expression using both Sin3b and NuRD repressor complexes. To obtain direct validation, we carried out ablation experiments of the two co-repressor complexes. Hence, depletion of Sin3b (Fig. 7J,K), CHD4 (Fig. 7L,M), HDAC1 (Fig. 7N,O) and HDAC2 (Fig. 7P) enhanced the expression of CDX2 at both transcript and protein levels.

Further, we wanted to address the physiological significance of BLM recruitment to the CDX2 promoter. We focused on ten paired samples from the Indian colon cancer patients. ChIP experiments revealed that the recruitment of BLM to the CDX2 promoter was lower in the cancerous regions compared to the adjacent normal regions (Fig. 8A). This indicates that in colon cancer patients increasing levels of CDX2 expression correlated with the absence of BLM from the promoter of this homeobox gene.

Next, we wanted to determine whether the increased levels of the DDSMs in the colon cancer patients correlated with the changes in the expression levels of their common target, BRCA1. We found that the transcript levels of BRCA1 decreased in the colon cancer patient tissue samples (for both stage I+II and stage III+IV) compared to their respective matched normal tissue controls (Fig. 8B). Analyzing only the cancerous tissues, we found that the levels of BRCA1 were significantly decreased in stage III+IV compared to their stage I+II counterparts. Western blot analysis (Fig. 8C,D) and immunohistochemistry (Fig. 8E,F) showed that BRCA1 protein levels were decreased in cancerous sections (across all stages) compared to their adjacent normal control. Finally, using the TCGA dataset, we observed that in the cancerous colon tissues, a negative correlation exists between the overexpression of the six DDSMs and the reduction of BRCA1 transcripts (Fig. 8G), thereby revealing the pathophysiological significance of the existence of these miRs.

Several miR signatures have been proposed to be predictive of the evolution of CRC. These include (1) a five miR signature identified through bioinformatics and subsequently validated in the tissues from two cohorts of patients (Ozawa et al., 2018); (2) an eight miR signature identified by three independent miR profile analyses that predict the recurrence of tumors in stage II and III patients (Kandimalla et al., 2018); (3) a four miR signature related to the relapse after curative surgery (Grassi et al., 2018); (4) a three miR signature that predicts both distant metastasis and hepatic recurrence (Coebergh van den Braak et al., 2018); and (5) a 16-miR signature that serves as a prognostic biomarker for stage II and III patients (Jacob et al., 2017). Distilling the results from these studies indicated that there was hardly any overlap in these lists of upregulated miRs in the different cohorts. Apart from miR-182, none of the DDSMs were present in any of the above studies. Such diverse results are possibly a consequence of the fact that multiple miRs can regulate the same pathway.

This study identified the CDX2 transcription factor as a common regulator of the DDSMs in colon cancer cells. A lot of studies have documented an altered pattern of CDX2 in CRC, resulting in a decreased or heterogenous expression, which correlated with compromised cell differentiation and adverse prognosis (Baba et al., 2009; Balbinot et al., 2018; Brabletz et al., 2004; Dalerba et al., 2016; Pilati et al., 2017). Several mouse models have supported the idea of tumor suppressor activity played by this homeobox gene as its reduction or loss of function facilitates the development of genetically- and chemically-induced intestinal cancers (Aoki et al., 2003; Balbinot et al., 2018; Bonhomme et al., 2003; Hryniuk et al., 2014; Sakamoto et al., 2017). This notion was further reinforced in cellular models by the inverse correlation between CDX2 levels and cell aggressiveness (Gross et al., 2008; Mallo et al., 1998; Yu et al., 2019; Zheng et al., 2017). However, increased CDX2 has also been reported in CRC (Witek et al., 2005), especially associated with short amplifications overlapping the gene locus and conferring a context-dependent oncogenic activity to this homeobox gene (Salari et al., 2012). The context-dependent effect is further illustrated in cases of abnormal ectopic expression outside the gut where CDX2 drives the formation of precancerous intestinal-type metaplasia mainly in foregut-derived organs (reviewed by Chawengsaksophak, 2019) and also promotes leukemogenesis in the hematopoietic lineage (Galland et al., 2021; Vu et al., 2020). Interestingly, we showed here that increasing CDX2 levels in colon cancer cells embedded in a growth factor- and matrix-rich environment stimulates tumor growth when grafted in NOD/SCID mice impaired for both T and B cells, unlike the less immunodeficient nude mice that only have a reduced number of T cells (Gross et al., 2008). However, the possibility exists that the effect of CDX2 expression may also regulate other factors that can also affect the process of tumour growth sustenance.

Although we show here that CDX2 stimulates DDSMs in colon cancer cells, the question remains as to why these miRs are not expressed in the normal intestinal epithelium? This could reflect the context-dependent molecular functions of CDX2, as besides its inductive transcriptional activity, it also has permissive effects by maintaining chromatin regions that are open and accessible to other sets of transcription factors that could be differentially expressed in the nucleus of malignant versus healthy cells (Saxena et al., 2018; Verzi et al., 2013). As CDX2 is a bivalent gene, it is not surprising to find BLM being recruited with its promoter together with HDAC1- and HDAC2-containing Sin3b and NuRD repressor complexes in the normal colonic mucosa and also in colon cancer cell, such as HCT116. The present study conducted on the CDX2 promoter in HCT116 colon cancer cells suggests that BLM can also play the role of an adaptor protein in unstressed cells, and that DNA damage leads to BLM redistribution to the site of damage and to its release from the CDX2 locus. Interestingly, an adaptor function has already been attributed to BLM whereby it helps in the interaction of c-Jun and c-Myc oncoproteins to their common E3 ligase, Fbw7α, and facilitates the degradation of the substrates (Chandra et al., 2013; Priyadarshini et al., 2018).

In an effort to understand how DDSMs carry out their biological functions in the cells, we have identified and validated common targets with known functions in DNA damage recognition, signaling and repair (i.e. the classical DDR response). Although four targets, BRCA1, ATM, Chk1 and RNF8, were initially validated, further functional and clinical studies focused on BRCA1 showed consistent effects on DNA repair, DNA damage and invasion using gain- and loss-of-function studies, and cell-based assays. It is interesting to note that BRCA1 and ATM are targeted by multiple miRs. For example, BRCA1 is downregulated by miR-182 (which is also one of the DDSMs) (Moskwa et al., 2011), miR-1255b, miR-148b*, miR-193b* (Choi et al., 2014), miR-146a, miR-146b-5p (Garcia et al., 2011) and miR-498 (Matamala et al., 2016). Similarly, ATM is targeted by miR-421 (Hu et al., 2010), miR-101 (Yan et al., 2010) and miR-203 (Zhou et al., 2014). As demonstrated previously (Moskwa et al., 2011), we also found that the miR-182-mediated suppression of BRCA1 impeded DNA repair, which led to a greater amount of residual DNA damage in miR-182-expressing cells compared to the control cells. Further analogous to the results in breast tumor lines (Moskwa et al., 2011), we also found that BRCA1 is targeted by miR-182 in colon tumor cells and patient tissues. The different miRs reported in the literature for ATM and BRCA1 could reflect specific types of DNA damage and/or different regulatory factors involved in their inductive mechanism(s). We also believe that the need for several miRs within the DDSM signature illustrates the relevance for appropriately controlling such an important process as DDR. The DDSMs bind to neighboring yet discrete sites on targets such as BRCA1 and possibly carry out cooperative repression in vivo (Broderick et al., 2011; Grimson et al., 2007; Saetrom et al., 2007). This also might be the reason why the DDSMs probably have neighboring non-canonical binding sites in the 3′ UTR of the target genes, such as BRCA1.

miRNA-miRNA interactions have been implicated in potential mutual regulatory patterns (Guo et al., 2012). miRNA-miRNA synergistic networks via co-regulatory functional modules have also been implicated in pathological conditions (Xu et al., 2011) and specifically in breast cancer (Cilek et al., 2017). We believe that a similar inter-regulatory network also possibly exists for the DDSMs. Hence, when one of the DDSMs was inhibited, the phenotypes associated with neoplastic transformation and cancer progression were cancelled out to a large extent.

This study provides evidence that DDSMs are biologically active moieties present in colon cancer patient tissue samples. Their overexpression in colon cancer patients correlated with decreased BRCA1 expression. Compared to matched normal tissues, BRCA1 is decreased in all stages of the cancerous tissues and also negatively correlates with cancer progression. The fact that high DDSM expression is associated with decreased probability of survival adds to the mechanistic reasons why the decrease in BRCA1 and ATM levels in colon cancer tissues correlates with reduced overall survival of the patients (Bai et al., 2004; Grabsch et al., 2006; Wang et al., 2018). If DDSMs respond to the extent of DNA damage and have a role in tumor growth sustenance, it may be possible to hinder tumor progression by inhibiting the miRs. Indeed, we showed here using two different experimental strategies that DDSM inhibition led to the regression of tumors. Similar approaches to target tumors using either RNA aptamers or chemical ligands have also been reported (Shu et al., 2014). Other complementary approaches, such as the usage of miRNA sponges, miRNA masking, antisense oligonucleotides or small molecule inhibitors (Hernández et al., 2018), could also be alternative methods to reach the same objective. The low levels of BRCA1 and ATM in cells having overexpressed DDSMs indicates defective DNA damage response and homologous recombination pathways. Hence, we suggest that colon cancer patients with increased DDSMs in tissues could be treated with PARP inhibitors, such as olaparib and veliparib, as already proposed in certain studies (Clark et al., 2012; Davidson et al., 2013; Wang et al., 2017).

In conclusion, the identification of DDSMs has importance as it presents an entire workflow for a colon cancer specific miR signature, whereby the identity of a common upstream regulator (CDX2) and the downstream effectors (DDR proteins such as BRCA1, ATM, Chk1 and RNF8) have been elucidated and validated in both preclinical mice models and patient samples (Fig. 8H). Further, this study opens up the possibility of the usage of the DDSMs as potential prognostic biomarkers of colon cancer and as targets for therapeutic miR inhibition, all attractive avenues of future research.

The antibodies used in the study are listed in Table S9, the recombinants listed in Table S10 and the reagents used are listed in Table S11. Information about the mammalian expression vectors of CDX2 used in this study is featured in Table S10 and has been described by Balbinot et al. (2017). pGEX4T-1 CDX2 was obtained by cloning full-length CDX2 into BamH1 and EcoR1 sites of pGEX4T-1. The DDSM promoter sequences with CDX2 binding sites are depicted in Tables S2-S5. The sequence containing the CDX2 binding site in the miR-96/182/183-5p promoter was cloned into the pGL3 vector using KpnI and HindIII to generate the pGL3-miR-96/182/183-5p wild-type promoter luc recombinant. BRCA1 coding and 3′ UTR sequence with four identified DDSM binding sites are depicted in Table S8. The BRCA1 3′ UTR sequences containing the miR-183 binding sequences were inserted into the KpnI and HindIII sites of the pGL3 vector to generate pGL3-BRCA1 3′ UTR miR-183 wild type. All site-directed mutagenesis was carried out using a QuikChange II XL Site-Directed Mutagenesis Kit. The siRNA and shRNA sequences used were as follows: CDX2 (5′-AACCAGGACGAAAGACAAAUA-3′); BLM (5′-AGCAGCGAUGUGAUUUGCA-3′); BRCA1 (5′-UCACAGUGUCCUUUAUGUA-3′); CHD4 (5′-CCCAGAAGAGGAUUUGUCA-3′); HDAC1 (5′-GGCUCCUAAAGUAACAUCAUU-3′); HDAC2 (5′-CCACCAUGCUUUAUGUGAUUU-3′); and Sin3b (5′-AGGCUGUAGACAUCGUCCA-3′). Information for miR mimics and inhibitors (used for DDSM overexpression and shutdown) is featured in Table S11. All primers in the study have been listed in Table S12.

miRNA expression levels of 332 colon tumor samples and eight normal samples were downloaded from the Genomic Data Commons (GDC) data portal (TCGA COAD) using the miR quantitation file and files with extension FPKM.txt.gz, respectively on 27 April 2017. The clinical information for all these samples was also downloaded from the GDC data portal. For the meta-analysis, miRNAs that were lowly expressed were first filtered out, followed by removal of the entries that were not expressed in at least half of the samples. DESeq2 (v.1.29.8) from R package was used to normalize the count values and then differentially expressed miRNAs compared to the normal samples were identified after performing a Wald statistical test. The miRNAs with a cutoff of log2fold-change (FC)>1.5 and adjusted P<0.05 were considered significant. The heatmap of the 62 upregulated miRNAs were plotted using R package gplots and clustered based on the group average clustering method. Further, we identified miRs upregulated across all stages of cancer compared to normal samples. One-way ANOVA was used to identify the significantly upregulated miRNAs consistently across the tumour samples.

For individual miR analysis, the log₂(x+1) transformation was carried out on the miR levels and gene expression values. The expression of these miRs in different stages of colon cancer compared to the normal samples was then examined by comparing the means of the expression values across different stages. A Kruskal–Wallis test was carried out in SPSS v.24 to compare the means of the miR expression across stages with that in the normal samples. The number of patients (n) in each classification was as follows: normal, 8; stage I, 67; stage II, 166; stage III, 119; stage IV: 61.

For Kaplan–Meier analysis, the expression value in each tissue sample was subtracted from the average value of normal samples. These values were then used to calculate the risk score of the six miRs found to be significantly upregulated in TCGA samples, as described previously (Ji et al., 2018). Briefly, regression coefficients (β) of the individual miRNA were determined by Cox regression analysis. The risk score was calculated for each patient using the formula: (β_miR29a-5p* expression value of miR-29a-5p)+(β_miR29a-3p* expression value of miR-29b-3p)+(β_miR96-5p* expression value of miR-96-5p)+(β_miR182-5p* expression value of miR-182-5p)+(β_miR183-5p* expression value of miR-183-5p)+(β_miR335-5p* expression value of miR-335-3p). The tissue samples with a risk score more than or equal to the 75% quartile or 0.9618 were grouped as high-risk samples (n=78), whereas the patients with a risk score less than or equal to the 25% quartile or −0.3654 were grouped as low-risk samples (n=76). The overall survival curve was plotted using Kaplan–Meier analysis in SPSS followed by a log-rank test to detect the significant difference between the high and low risk group of patients. Spearman correlation was carried out with 207 patient samples for which both the transcriptome and miR expression datasets were available.

All samples (detailed in Table S13) pertaining to the Indian cohort were obtained from the All India Institute of Medical Science (according to the Institutional Human Ethics Committee approval number RP-23/2017) after obtaining informed consent from all subjects. All experimental work on these samples was carried out at the National Institute of Immunology (according to the Institutional Human Ethics Committee approval number IHEC#92/17 and Institutional Bio-Safety Committee approval number IBSC/VKN/2014/63). All samples obtained were paired, and the adjacent normal tissues were excised 7-10 cm from the periphery of the tumor. Each tissue sample was graded, subjected to routine histology and Haemotoxylin and Eosin staining, and then the core of the tumour was used for RNA/protein/immunohistochemical analysis. Patients in polyp, stages I and II were combined together and stages III and IV were combined together. The samples were acquired in two phases – phase 1 (n=40) and phase 2 (n=14). In phase 1, out of 40 patients, 26 patients were in stage I+II, whereas 14 patients were in stage III+IV. In phase 2, out of 14 patients, five patients were in stage I+II, whereas nine patients were in stage III+IV. Significantly altered expression of miRs, BRCA1 mRNA and BRCA1 protein (estimated by both western blot and immunohistochemistry) were identified in the tissues of the Indian cohort.

All animal studies were carried out at the National Institute of Immunology according to an approved animal ethics protocol (Institute Animal Ethics Committee approval reference number: IAEC#398/15). The following animal studies were carried out: (1) tumorigenic potential of HCT116 BLM^−/− cells expressing miR inhibitors using a subcutaneous model; (2) tumorigenic potential of HCT116 BLM^−/− cells that were challenged with nanoparticle-coated miR inhibitors in a subcutaneous model; (3) tumorigenic potential of HC1/HW2 cells using a subcutaneous model; (4) tumorigenic potential of TG8/TW6 cells using a subcutaneous model; and (5) tumorigenic potential of TG8/TW6 cells using an orthotopic model. In all subcutaneous models (1-4) ∼2 million cells were resuspended in fetal bovine serum. These cells were injected subcutaneously into healthy 8-week-old male or female NOD SCID mice along with matrigel. In the TG8/TW6 models (4 and 5) the mice were additionally treated with Dox (10 mg/ml/kg body weight) by oral gavage every day throughout the experiments. In all the subcutaneous models (1-4), tumor formation started after 7-10 days. In the case of study 2, a nanoparticle-mediated delivery system was used to deliver miR inhibitor control, miR inhibitor 29a-5p or miR inhibitor 96-5p directly to the base of the tumors four times every third day (further details of the delivery system are described below in a separate section). In the case of the orthotopic model (5), 50,000 TG8/TW6 cells were used and were implanted into the cecal wall. In study 5, at the end point, whole-body imaging was performed for the mice using an in vivo imaging system (PerkinElmer) to check the expression of GFP and thereby determine the invasive potential of TG8/TW6 cells. For all the above models, at the endpoint, the mice were sacrificed by cervical dislocation. The excised tumors (in case of subcutaneous models, 1-4) were imaged and measured, and used for lysate preparation (using RIPA), RNA extraction (using Trizol LS) and immunohistochemistry. The immunohistochemistry staining is presented as an H-score (Rajarajan et al., 2020).

All pre-existing cell lines (Table S11) were maintained as described in the original publications or as recommended by the suppliers. Immortalized cell lines from BS patient GM03509 were complemented with either GFP-BLM (Clone 4.3.4) or with GFP (Clone 100) (Kharat et al., 2016). Immortalized cell lines from another BS patient GM08505 were also used (Hu et al., 2001). Cells from the BS patients were derived from the respective patients' skin fibroblasts. HC1/HW2 cells were generated by stably transfecting HCT116 cells with pCB6 (for HC1) or pCB6-Flag₂-mCDX2 wild type (for HW2) and selecting with G418. The cells were grown in DMEM plus 10% FBS plus G418 (1 mg/ml) plus antibiotics. To generate the HCT116 BLM^−/− Inhi control cells, pLenti-III-mir-Off control vector (Abm Inc.) was used. The lentivirus was generated by using Lenti-X HT packaging mix in Lenti-X 293T. To obtain HCT116 BLM^−/− Inhi-29a-5p and HCT116 BLM^−/− Inhi-96-5p lines, commercial lentivirus particles were used (Abm Inc.). BLM^−/− cells were plated in a six-well cluster and transduced with the three different lentiviral particles. Transduction was carried out with 2 μg/ml polybrene. Medium was changed 24 h post-transduction. For selection, 1 µg/ml puromycin was added to the cells. The transduced cells were grown in the presence of 1 µg/ml puromycin for stable line generation for 7 days, after which the clones were analyzed for the expression of the two miRs. ML216 (12.5 µM) treatment was carried out for 24 h prior to making the chromatin for the ChIP assays. For DNA damage-dependent experiments, cells were exposed to a particular IR (3 Gy) or a range of IR indicated in the specific experiments. Lysates were made, or RNA extracted, 1 h or 6 h post-IR exposure. All cells used tested negative for mycoplasma contamination.

RNA immunoprecipitation was carried out in HeLa pREV and Ago2 cells according to published protocols (Keene et al., 2006). Cells were scraped in PBS and resuspended in polysome lysis buffer [100 mM KCl, 5 mM MgCl_2, 10 mM HEPES (pH 7.0), 0.5% NP-40, 1 mM DTT, 100 units/ml RNase Out, 400 µM vanadyl ribonucleoside complexes (VRC), Protease inhibitor cocktail supplemented with RNase inhibitor and protease inhibitors]. The lysates (2 mg) were used to set up RNA immunoprecipitations with either Protein G-bound GFP antibody (2 µg/immunprecipitation) or with anti-Flag beads (4 µl/immunoprecipitation). The immunoprecipitations were carried out for 4 h on an end-to-end rotor at 4°C, after which beads were pelleted at 300 g for 5 min and washed with ice-cold NT2 buffer three or four times. Beads were resuspended in 100 µl of NT2 buffer [50 mm Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM MgCl₂ and 0.05% NP-40] supplemented with 30 µg of Proteinase K to release the RNP components. This mixture was incubated at 55°C for 30 min. Thereafter, 200 µl of Trizol were added directly to the beads and RNA was isolated. This was followed by cDNA synthesis and qPCR to determine the enrichment of miRNA bound with Argonaute 2 protein. The main steps of traditional immunoprecipitations were similar to above, except 1× PBS plus 0.1% NP-40 was used as the buffer to make up the volume and carry out washes. Post-immunoprecipitation, the bound proteins were run in SDS-PAGE gels to determine their co-immunoprecipitating partners.

Cells were transfected with 20 nM of either miRNA inhibitors or miRNA mimics specific to the respective miRs. Both miR inhibitors and miRNA mimics were commercially purchased and their details are described in Table S11. Lipofectamine 2000 was used for the transfections in a 1:1 ratio with the amount of inhibitor or mimic being used. Control miRNA inhibitor or control miRNA mimic at the same concentrations were always used in parallel. Transfections took place for a 6-h time period. Next, 36 h post-transfection, either RNA was isolated or lysate was prepared using RIPA buffer [1 mM Tris HCl (pH 7.8), 150 mM NaCl, 2% Triton X-100, 1% (w/v) Sodium deoxycholate and 0.1% (w/v) SDS supplemented with 1× protease inhibitor cocktail (PIC) and 1 mM phenylmethylsulfonyl fluoride]. Transfections involving plasmids were carried out using the respective plasmids in six-well cluster plates for 48 h. All siRNA transfections were carried out using 200 pmol of the respective siRNAs for 60 h. For shSin3b induction, cells transfected with pTRIPZ shSin3b were treated with Dox (1 µg/ml) for 48 h. Corresponding siRNA or shRNA controls were always used in parallel for all the ablation experiments.

For each EMSA reaction the radiolabelled substrate (10⁴ cpm) was incubated with 500 ng of recombinant CDX2 protein in a binding buffer [10 mM Tris (pH-7.5), 50 mM KCl, 2.5 mM MgCl₂., 0.5 mM DTT and 4% glycerol] for 15 min at 4°C. Poly dI-dC (1 μg/μl) was added to the reaction mixture to prevent non-specific binding. When ‘supershift’ was desired, anti-CDX2 antibody (1 µg/reaction) was added and incubation continued at 37°C for another 15 min. In certain reactions, 1000× fold-excess cold competitor was also added to confirm the specificity of the assay. After the reactions were complete, loading dye (5× TBE, 10% glycerol, 10% bromophenol blue, 1% xylene cyanol and autoclaved H₂O) was added in all the tubes. Samples were then loaded onto a 6% native PAGE gel that had already been pre-run in TBE buffer at 20 mA and 300 V for 30 min. Sample without CDX2 was used as control for the EMSA reactions. After the run was completed, the gel was dried at 65°C for 1 h and exposed overnight, followed by autoradiography.

Total RNA was isolated from an asynchronously growing BLM isogenic pair of cells – (a) GM03509 GFP-BLM 4.3.4/GM03509 GFP (b) Clone 100 and GM08505 GFP-BLM/GM08505 GFP. RNA was extracted using Trizol and the isolated RNA was used for library preparation using an Illumina Small RNA sample preparation kit v1.5 according to the manufacturer's instructions. The total RNA (700-800 ng) was ligated to 3′ and 5′ RNA adapters. The ligation products were reverse transcribed using Superscript II Reverse Transcriptase and amplified with 12 cycles of PCR. The PCR products constituting the small RNA cDNA libraries were resolved on a 6% Novex TBE PAGE gel and ∼150 bp fragments were excised. The library was eluted from the PAGE gel and analyzed on an Agilent 2100 Bioanalyzer using a DNA high sensitivity kit (Agilent Technologies, USA). Sequencing of miRNA libraries (∼150 bp fragments) was performed using an Illumina GAIIX sequencing platform for 36 cycles. CLC genomic software was used to quantitatively determine the levels of the differentially expressed miRNAs in the two isogenic pairs. With the help of this software, adaptor trimming was also performed. The remaining sequence was mapped with the known miRNAs in the miRBase database (www.mirbase.org). Subsequently, the mapped miRNAs were normalized to get transcripts per million. Further analysis was carried out to determine the miRNAs that had expression levels that were either increased or decreased by the presence of BLM. Only those miRNAs that showed a change above or below two-fold and P≤0.05 were chosen for further analysis.

In order to determine the transcription factors involved in the regulation of the miRNAs, initial in silico analysis was carried out using the database called ChIPBase (http://rna.sysu.edu.cn/chipbase3/index.php). The region up to 5 kb upstream of the TSS was considered as the promoter region of the miRs. Using the ChIPBase v3.0 database, a list of transcription factors binding to the promoter regions of the DDSMs was determined. Only the transcription factors binding to at least four out of the six DDSM promoters are depicted in Table S1.

From putative targets for each of the miRs (Table S6), the common targets of all DDSMs were predicted (Table S7). Prediction of the binding sites of the DDSMs in the 3′ UTR of BRCA1 (Table S8) was performed using the miRanda 3.3a (https://bioweb.pasteur.fr/packages/pack@[email protected]). The potential gene targets for each of the miRNAs were predicted and then plotted using jvenn software (Bardou et al., 2014). Pathway analysis was carried out using the Reactome, KEGG, Wiki Pathways and GSEA online databases, and the significant pathways with P<0.05 were selected.

The cationic polymer (TAC6) was used for the in vivo delivery of miRNA inhibitors (Yavvari et al., 2019). miRNA inhibitors (10 µl of 20 µM) were mixed with TAC6 polymer (final volume 100 µl at 1 mg/ml) and incubated for 20 min at room temperature. Complexes were then mixed with sodium aspartate (final volume 10 µl at 1 mg/ml) for 10 min. The nanogels created were diluted with PBS. Each mouse was given a dose of 200 ng miRNA (50 µl of nanogels). A total of four doses were used.

The BLM immunoprecipitate was electrophoresed on SDS-PAGE and stained by Coomassie. Each lane of the gel containing BLM-interacting proteins was subjected to mass spectrometry. Briefly, each lane was cut out from the gel separately into smaller pieces. Coomassie-stained gel pieces were de-stained using 25 mM ammonium bicarbonate and 50% (v/v) acetonitrile solution. Subsequently, they were treated with 0.1 M TCEP for 45 min at 37°C, followed by 0.5 M iodoacetamide for 1 h at 37°C. Overnight tryptic in-gel digestion was then carried out (trypsin: protein ratio of 1:100). The next day, peptides were recovered. The pH of the supernatant was acidified (∼pH3) using trifluoroacetic acid. The supernatant was dried in a SpeedVac. Resuspension was carried out in 5% acetonitrile and 0.1% formic acid. Desalting was carried out using ZIP TIP (C18 P-10, Millipore). The eluted peptides were dried using a SpeedVac. The peptides were finally resuspended in 5% acetonitrile and 0.1% formic acid.

All mass spectrometry experiments were performed using an EASY-nLC system (Thermo Fisher Scientific) coupled to an LTQ Orbitrap-Velos mass spectrometer (Thermo Fisher Scientific) equipped with a nano-electrospray ion source. A 10-cm PicoFrit Self-Pack microcapillary column (New Objective) was used to resolve the peptide mixture and the peptides were eluted. The LTQ Orbitrap-velos was operated using the Top20 CID (High/High) data-dependent acquisition mode with a full scan in the Orbitrap and a tandem mass spectrometry scan in the CID. The target values for the full scan mass spectrometry spectra were set at 0.5×10⁶ charges with a maximum injection time of 300 ms and a resolution of 60,000 at m/z 400. Obtained spectra were queried against the human UniProt database. The precursor and fragment mass tolerances were set at 10 ppm and 0.8 Da, respectively. Proteome Discoverer 1.3 was used as the search algorithm, with oxidation of methionine and carbamido-methylation of cysteine as static modification. All peptide-spectrum matches (PSMs) were identified at a 1% false discovery rate. For peptide identification, a peptide posterior error probability threshold of 0.01 was specified. A comprehensive and nonredundant list of all human proteins along with their UniProt ID and functional classification was generated using the UniProt database. The same Sin3b and CHD4 peptides were detected in the BLM immunoprecipitates three times.

All quantifications are presented as mean±s.d. except Fig. 1B where it is median±range.

The number of replicates are mentioned in the figure legends. The P values or calculated probability is as follows: *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001, and n.s. indicates that the result is not significant. The statistical analyses employed for every experiment are shown in Table S14.

We thank Dipanjan Chowdhury (Dana-Farber Cancer Institute, USA), Nathan Ellis (The University of Arizona Cancer Center, USA), Gregory David (New York University School of Medicine, USA), Jeff Wrana (University of Toronto, Canada) and Joan Massague (Sloan Kettering Institute, USA) for recombinants; Nathan Ellis (The University of Arizona Cancer Center, USA), Annick Harel-Bellan (University of Paris-Saclay, France), Bert Vogelstein (Johns Hopkins Medicine, USA) for cells; Aasheesh Srivastava (IISER Bhopal) for polymer; Vaibhav Jain for help with bioinformatic analysis; and Kunal Dhall for clinical data compilation. S.S. and A.M. are grateful to the Department of Biotechnology (DBT), Government of India, for a generous infrastructure grant for the establishment of the Next Generation Sequencing Core Facility in the National Institute of Immunology.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.258601