The IRE1α Endonuclease Plays a Dual Role in Regulating the XBP1/miRNA-34a Axis and PD-1 Expression within Natural Killer Cells in Hodgkin Lymphoma Free

It is established that PD-1⁺ natural killer (NK) cells are functionally impaired in classical Hodgkin lymphoma (HL) [1]. However, NK cell dysfunction remains poorly characterized in this disease. NK cells are an important subset of innate effector cells that possess the unique ability to lyse malignant cells without previous sensitization [2, 3]. Until recently, attention on immune effector cell function in classical HL has largely focused on cytotoxic T lymphocytes (CTL), in part because relative to CTL the proportion of NK cells within malignant lymph nodes are very modest. However, whereas only a small fraction of CTL will express the relevant T-cell receptor capable of recognizing a malignant cell in an antigen-specific manner, the majority of NK cells are able to exert antitumoral cytotoxicity. Furthermore, their cytotoxic capabilities allow them to kill tumor cells even at relatively low ratios [4]. For these reasons, NK cells have recently attracted attention for their potential use as immune-based therapies [5].

A well-characterized immune-evasion strategy in HL includes copy number gain of Janus kinase signaling and Epstein-Barr virus-driven gene amplification of programmed cell death protein ligand 1 (PD-L1) and PD-L2 ligands in the malignant Hodgkin Reed-Sternberg (HRS) cells [6, 7]. Interaction between PD-L1/PD-L2 and their cognate receptor PD-1 on immune effector cells has been shown to dampen antitumor activity [8]. Although blockade of the PD-1 axis with anti-PD-1 antibodies has proven clinical utility in relapsed/refractory HL [9, 10], relapses frequently occur, indicating the need to mobilize additional immune effector mechanisms. Even though HL diseased lymph nodes contain few NK cells, several lines of evidence suggest that HL is a highly attractive disease to harness peripheral NK cells to the lymphoid circulation. In the missing-self hypothesis, NK cells preferentially lyse cells with reduced or lost expression of major histocompatibility complex class I (MHC-I) molecules while sparing cells with intact MHC-I expression [11]. Notably, the HRS cells in HL have frequently lost MHC-I on their cell surface [12]. It is known that NK cells are dysregulated in HL, and that PD-1 blockade partially abrogates NK cell dysfunction [1] and appears to synergize with bi-specific NK cell engagers [13]. Among relapsed/refractory patients with HL, the number of circulating NK cells is associated with response to PD-1 blockade [14], which is associated with a peripheral immune signature that includes an abundance of activated NK cells [15]. Intriguingly, in responsive patients with HL receiving PD-1 blockade, shortly after treatment the numbers of peripheral blood circulating NK cells are reduced, potentially consistent with their migration to the lymphoid circulation [16]. Furthermore, although the transcriptional mechanisms that regulate PD-1 expression in NK cells are of great importance and interest, they are not yet well understood. Elucidating these mechanisms may enable novel NK cell-driven immunotherapeutic strategies to be pursued.

Inositol-requiring protein 1α (IRE1α) is a serine/threonine-proteinase/endoribonuclease stress sensor that operates within the unfolded protein response system. Under endoplasmic reticulum (ER) stress, IRE1α undergoes oligomerization and subsequent autophosphorylation that triggers its endonuclease activity. The latter enables processing of its principal substrate, X-Box Binding Protein 1 (XBP1) mRNA, via excision of a 26-nucleotide intron and generation of the active form, the transcription factor XBP1-spliced (XBP1s) [17]. The IRE1α pathway is a highly evolutionarily conserved cell signaling pathway with multiple roles in both innate and adaptive immune cells including plasma cells, dendritic cells, eosinophils, macrophages, T cells, and NK cells [18‒25]. Within these immune cell subsets, the functional role of the IRE1α pathway is highly cell context-specific.

Additionally, it has been established in cell-free systems that the IRE1α endonuclease recognizes a consensus sequence (CUGCAG) motif that enables it to cleave selected microRNAs (i.e., miR-17, miR-34a, miR-96, and miR-125b) during ER stress [26, 27]. MicroRNAs (miRNAs) function in part by targeting the 3′ untranslated regions (3′UTRs) of messenger RNAs (mRNAs) for degradation and/or translational repression to “fine-tune” expression of various proteins. Of particular interest within an immune context is miR-34a-5p, which binds to the 3′UTR of PD-L1 in the malignant cells of various solid malignancies and blood cancers [28‒30]. The regulation of the miR-34a-5p/PD-L1 axis by diverse stimuli such as the p53 pathway and Epstein-Barr virus within a range of malignant cell types strongly implicates it as an important regulatory hub that can be manipulated for therapeutic benefit [28‒31].

Interestingly, miR-34a-5p is also expressed by a range of immune effector cells including dendritic cells, macrophages, mast cells, B cells, and T cells [32‒35]. Importantly, miR-34a-5p is a key regulator of multiple inhibitory/activating receptors and chemokines in NK cells [36]. Recently, it has been shown that the IRE1α pathway mediates NK cell responses toward viral infections and murine melanoma tumor models in vivo, as well as cytotoxic function and proliferation of human NK cells from healthy blood donors [24, 25]. To our knowledge, there are minimal data on the functional role of the IRE1α pathway in primary NK cells (pNK cells) from patients with human malignancies, and the role of a potential IRE1α-miR-34a-5p axis in NK cells remains unexplored. In the current study, we sought to investigate the role of the IRE1α pathway on effector function and PD-1 expression in NK cells of patients with HL. Results point to IRE1α endonuclease playing a dual role in regulating the XBP1/miRNA-34a axis and PD-1 expression within NK cells, which is disrupted in patients with HL.

NK cell lines used were SNK10 (chronic active Epstein-Barr virus) and KHYG-1 (NK cell leukemia). Target cell lines were HDLM2 (HL) from a patient with poor-risk (elderly, advanced stage disease), KMH2 (HL) (chemo-refractory disease), and K562 (HLA-deficient erythroleukemia). Effector:Target (E:T) ratios used in experimental assays were 1:1. NK cell lines were typically stimulated for 2 h (K562) or 8 h (KMH2 and HDLM2). All RT-qPCR assays to measure gene expression related to activated NK cells were performed with NK cells following isolation from the cocultures with target cancer cells (NK cell isolation kit, Miltenyi Biotec).

The IRE1α pathway was activated by treatment with thapsigargin (TG) and was inhibited by using small-molecule inhibitors such as 8-formyl-7-hydroxy-4-methylcoumarin (4µ8c) or 6-bromo-2-hydroxy-3-methoxybenzaldehyde (6-bromo). Knockdown of IRE1α expression was performed using Dicer-substrate short-interfering RNA (Integrated DNA technologies) as outlined in the online supplementary material (for all online suppl. material, see https://doi.org/10.1159/000536044).

Putative target sites in the 3′UTR region of XBP1 and PD-1 were predicted for IRE1α-regulated miRNAs (miR-17, miR-34a, miR-96, and miR-125b) using miRanda algorithm. Mature miRNA sequences for all four miRNAs were downloaded from the miRBase database (Release 22) [38], and the XBP1 and PD-1 transcript sequences were retrieved from GENCODE (Release 21 Reference Human Genome GRCh38) [39]. Only predicted target sites with stringent parameters (free energy ≤−18 and miRanda score ≥140) were considered for further experimental validations (online suppl. Table S1). Functional validation of miRNA binding sites including the dual luciferase assay and transfection of miRNA mimics is outlined in more details in the online supplementary material.

The computational prediction of XBP1 binding sites was performed on regulatory regions of the PDCD1 (PD-1) gene found to be active in human NK cells (ENCODE; www.encodeproject.org/experiments/ENCSR808HWS/). We used MEME motif discovery program [40] for a de novo discovery of XBP1 binding sites on sequences enriched in regions experimentally determined by XBP1-ChIP assay and RNA-seq (as described elsewhere [41] and datasets available at ArrayExpress: E-MTAB-6327, E-MTAB-6894, and E-MATB-7104). We identified overrepresented DNA patterns (or motifs) in the XBP1 binding regions of the top 500 most upregulated genes in activated TH2 (high levels of XBP1s) versus naïve TH2 (889 peaks) and the top 500 most downregulated genes in activated TH2 versus naïve TH2 (595 peaks). Two putative XBP1 motifs were found: one canonical motif similar to the one described in JASPAR database (MA0844) and a novel variant XBP1 motif not yet described in any available database or publication. TAMO package [42] was used for the identification of putative XBP1 binding sites in the promoter regions of PD-1 gene. Only putative binding sites at least 70% similarity to canonical or variant XBP1 motifs were selected.

A luciferase reporter system (pGL3 basic vector, Promega) was used to validate the direct interaction of XBP1s and the predicted promoter regions of PD-1 as well as the individual XBP1s binding sites identified within PD-1 promoter regions. More details of methods are described in the supplemental material.

Pre-therapy blood samples from 14 patients with HL were collected before therapy (median age 37 years, range 16–73; F:M ratio 9:5; 6 stage I-II, 6 stage III-IV, 2 stage unknown; 10 nodular sclerosing, 1 lymphocyte rich, 3 classical HL otherwise not classified); and from 10 healthy participants (median age 37 years, range 25–48, F:M ratio 5:5). Peripheral blood mononuclear cells were isolated and cryopreserved, as described previously [43]. This study conformed to the Declaration of Helsinki, and informed consent was provided by all participants in accordance with participating hospitals/research institute Human Research Ethics Committee guidelines.

pNK cells were isolated from cryopreserved peripheral blood mononuclear cells and expanded as previously described [1]. Flow cytometry, NK cell migration, NK cell immune synapse (NKIS) formation, and cytotoxicity are outlined in the supplemental material. For PD-1 blockade experiments, NK cells were cultured for 72 h with 10 µg/mL of pembrolizumab^™ (or IgG4 control), prior to the addition of targets. At the completion of the culture, the KHYG-1 cells or pNK cells were collected and incubated with an equal number of targets. Assays were performed in triplicate. More details are provided in the online supplementary material.

All statistical analysis was done using GraphPad Prism 7.00 (GraphPad Software). Comparison of measurements was performed using paired 2-tailed t tests (except when stated otherwise in figure legends). Only p < 0.05 was considered significant.

To confirm that the IRE1α pathway can be activated in NK cells in response to blood cancer cell lines, we coincubated the NK cell-line SNK10 with the HRS cell-lines HDLM2 (MHC-I deficient) or KMH2 (MHC-I intact), and the NK cell-sensitive target leukemia cell line K562 (MHC-I and II deficient). Splicing of the transcription factor XBP1 to its active form XBP1s was used as the functional readout to indicate IRE1α-endonuclease activation. The mRNA expression of XBP1s in K562 activated SNK10 cells increased by at least 2-fold in comparison to unstimulated cells (shown in Fig. 1a). Similarly, a significant increase in XBP1s protein expression was observed in stimulated SNK10 cells (shown in Fig. 1b; online suppl. Fig. S1). However, the increase of XBP1s protein expression in SNK10 cells stimulated with both HL cell lines (KMH2 and HDLM2) was far more pronounced (up to ∼5–6-fold) than with K562 leukemia cells (∼2-fold; shown in Fig. 1b; online suppl. Fig. S1).

In addition to activating XBP1s splicing, it has been shown in cell-free systems that the IRE1α pathway also plays a role in the degradation by cleavage of a subset of CUGCAG motif-containing miRNAs including miR-34a [27]. Next, we sought to establish if degradation of miR-34a is operative within NK cells. First, we measured miR-34a-5p expression by qRT-PCR in SNK10 cells stimulated with these three different target cell lines: K562, KMH2, and HDLM2. This showed significant downregulation of miR-34a-5p in SNK10 cells after stimulation with all three targets cell lines (shown in Fig. 1c). To confirm the mechanism of IRE1α-mediated regulation of XBP1s and miR-34a-5p is conserved in cells other than NK cells, we tested IRE1α activation in HEK293T cells using a well-known ER stress inducer TG. Activation of the IRE1α pathway in HEK293T cells by TG downregulated miR-34a-5p expression, whilst XBP1s isoform was upregulated (shown in Fig. 1d). These results strongly suggest that IRE1α endonuclease activity plays a dual role in the control of expression of XBP1s and miR-34a-5p within NK cells.

Although miR-34a-5p is known to regulate multiple inhibitory/activating receptors and chemokines in NK cells [36], its potential role in regulating PD-1 expression in NK cells remains unknown. Similarly, although it has also been shown that miR-34a-5p can target XBP1 in neuronal cells [44], the role of this miRNA in NK cells is unexplored. Therefore, to examine this further, we looked for putative binding sites for the miRNAs in the 3′UTR of XBP1 and PD-1 mRNA. Next, we performed a computational analysis to identify putative target sites for all four miRNAs known to be regulated by IRE1α (miR-17, miR-34a, miR-96, and miR-125b) [27], in the 3′UTR region of XBP1 and PD-1 mRNAs. Adopting a stringent criterion for miRNA target site prediction (miRanda score ≥140 and free energy <−18 kcal/mol), we found that only miR-34a-5p had putative target sites for the 3′UTR of XBP1 and PD-1 mRNAs (online suppl. Table S1). Therefore, efforts were focused on validating binding of miR-34a to XBP1 (two target sites) and PD-1 (one target site) in NK cells.

To validate the three putative target sites of miR-34a-5p in the 3′UTR of XBP1 and PD-1 transcripts, we performed a dual luciferase reporter assay [45]. We tested the regulatory effect of miR-34a-5p mimics on the expression of Renilla Luciferase (Rluc) fused with the predicted target site found in the 3′UTR of XBP1 (shown in Fig. 2a, b) or PD-1 (shown in Fig. 2c) using HEK293T cells to host the dual luciferase expression vector. For this assay, Caenorhabditis elegans miR-67-5p (cel-miR-67-5p), with no putative binding sites found in XBP1 and PD-1 3′UTRs, was used as a negative control to confirm the specific repression of Rluc (carrying candidate target sites) by miR-34a-5p (shown in Fig. 2a–c). We then performed a rescue assay by quantifying the expression of Rluc carrying the miR-34a-5p target sites after treatment with miR-34a-5p mimics and rescue with miR-34a-5p antagomir (anti-miR-34a-5p) compared to negative control antagomir (anti-miR NC, a nontargeting negative control for antagomir assays that do not recognize any sequences in human transcriptomes including miR-34a-5p sequence). As expected, normalized Rluc signal was significantly reduced by the presence of miR-34a-5p mimics (plus anti-miR NC; shown in online suppl. Fig. S2) when compared to the rescue condition (plus anti-miR-34a-5p; shown in online suppl. Fig. S2). In addition, we performed a biotinylated miR-34a-5p/mRNA pull-down assay to confirm a significant enrichment of XBP1 and PD-1 mRNAs (shown in Fig. 2d).

To further confirm the regulatory effect of miR-34a-5p on XBP1 and PD-1 expression, we treated the PD-1 expressing NK cell-line KHYG-1 with miR-34a-5p mimics. There was a significant knockdown of XBP1 and PD-1 mRNA in miR-34a-5p mimics-treated KHYG-1 cells compared to negative control treatment (cel-miR-67-5p; shown in Fig. 2e). These results confirm that miR-34a-5p can suppress both XBP1 and PD-1 mRNAs in NK cells. To further confirm the association between miR-34a-5p and PD-1, we assessed the correlative association between miR-34a-5p and PD-1 in vitro by using the PD-1 expressing NK cell-line KHYG-1. PD-1^high and PD-1^low KHYG-1 cells were sorted with fluorescent-activated cell sorter. Relative quantities of miR-34a-5p and PD-1 transcripts in the two KHYG-1 populations were assessed by qRT-PCR. In agreement with our hypothesis of PD-1 repression by miR-34a-5p, we observed a significant downregulation of miR-34a-5p expression in PD-1^high KHYG-1 cells when compared to PD-1^low KHYG-1 cells (shown in Fig. 2f).

A time course experiment for IRE1α blockade using DsiIRE1 (Dicer-substrate RNA interference designed to knockdown IRE1α expression) assay showed a significant increase in miR-34a expression in NK cells that is inversely proportional to IRE1α expression in keeping with an IRE1α-negative regulation of miR-34a-5p. For this experiment, we again used the PD-1 expressing NK cell-line KHYG-1 to interrogate the role of IRE1α in the regulation of miR-34a and PD-1. First, we confirmed the inhibitory effect of DsiIRE1 on IRE1α mRNA with maximum knockdown of IRE1α mRNA observed at 12 h post-DsiIRE1 treatment (shown in Fig. 3a), although IRE1α knockdown lasted for 24 h post treatment. As expected, the two arms of miR-34a, miR34a-5p and miR-34a-3p, showed an inverted association with IRE1α mRNA expression levels (shown in Fig. 3b, c). The expression of both mature sequences of miR-34a hairpin (miR-34a-5p and miR-34a-3p) peaked at 12 h and remained significantly upregulated until 24 h post DsiIRE1 treatment. In contrast, the cells with IRE1α inhibited by DsiIRE1 demonstrated significantly reduced PD-1 mRNA expression at 12 h post treatment, which remained downregulated until 24 h post treatment (shown in Fig. 3d). Therefore, we further validated the effect of IRE1α on PD-1 protein expression using TG (IRE1α activator) and DsiIRE1 (IRE1α inhibitor) treatments in the KHYG-1 cells. Our experiments confirmed a significant reduction in PD-1 protein expression in the cells treated with DsiIRE1 at 12 h and more so at 24 h (shown in Fig. 3e). On the other hand, IRE1α activation by TG treatment showed a stimulating effect on PD-1 protein expression and this effect was significant even after 24 h in the KHYG-1 cells treated with DsiIRE1 compared to negative control DsiNC (shown in Fig. 3f). Furthermore, we also tested if IRE1α knockdown affects XBP1s expression. We found that XBP1s expression is significantly downregulated at 12 h post-DsiIRE1 treatment (followed by 2 h of TG activation; online suppl. Fig. S3A, B) when IRE1α is silenced. Moreover, XBP1s expression is positively correlated with IRE1α transcription (Pearson’s correlation coefficient = 0.98, p = 0.02; Fig. S3C), providing additional evidence that IRE1α transcription is directly associated with XBP1s splicing output.

We then investigated whether XBP1s putatively binds to the promoter and/or enhancer regions of PD-1 gene. Putative XBP1 target sites were identified using the canonical XBP1 binding motif as described in the JASPAR database (MA0844.1) [46] and a novel variant XBP1 motif that we found in a de novo discovery cohort using XBP1-ChIP assay data previously published [41] (shown in Fig. 4a; see Methods for details). The canonical XBP1 binding motif reported in JASPAR database is derived from published collections of experimentally defined transcription factor binding sites, whereas the novel variant XBP1 motif is not reported anywhere in the literature. Next, similar occurrences (at least 70% of similarity) of these two alternative XBP1 motifs were searched in active regulatory regions of PD-1 previously determined by ATAC-seq (Assay for Transposase-Accessible Chromatin using sequencing) in primary human NK cells (ENCODE; www.encodeproject.org/experiments/ENCSR808HWS/). Two canonical XBP1 target sites were identified in downstream regions of PD-1 gene, while several putative sites were found for the new variant XBP1 motif distributed across the upstream, intronic, and downstream regions (shown in Fig. 4b).

We selected three putative regulatory regions, upstream, intronic, and downstream, of the PD-1 gene where chromatin is accessible in human NK cells and putative XBP1 binding sites were predicted (shown in Fig. 4b). All three regulatory regions were amplified from genomic DNA extracted from KHYG-1 cells and cloned into the promoter of a luciferase gene in a PGL3 vector. HEK293T cells were transfected with PGL3 vectors containing the upstream (810 bp), intronic (437 bp), or downstream (280bp) regions. Transfected HEK293T cells were then treated with TG (positive control for XBP1s activation) or DMSO (negative control in which XBP1s is not expressed) and the luciferase activity measured using a luciferase reporter assay (compared to native PGL3 vector carrying basic promoter as background control). The results indicate that XBP1s can significantly promote the expression of luciferase gene carrying the upstream and downstream regions containing XBP1s binding sites found in active chromatin regions nearby PD-1 gene (shown in Fig. 4c).

We then tested a few selected putative XBP1s binding sites (highest similarity to XBP1s motifs) identified within each regulatory region for further validation. The most abundant XBP1s binding site found in these regions was the variant XBP1s motif (shown in Fig. 4b) with the canonical XBP1s motif found only at the downstream region and a more distal accessible region further downstream of PD-1 gene (shown in Fig. 4b). Specific putative XBP1s binding sites found in these regulatory regions were cloned into PGL3 vector. The luciferase expression measured in KHYG-1 cells (DMSO or TG treated) transfected with PGL3 vectors carrying candidate XBP1s binding sites revealed that the variant XBP1s binding sites (S1, S2, and S4) can enhance XBP1s-driven expression of luciferase gene (shown in Fig. 4d), while the canonical XBP1 binding sites (S5 and S6) were not able to elicit any significant change in luciferase expression (shown in Fig. 4e).

To investigate the translational relevance of our findings, our study was extended to pNK cells obtained from patients with HL, a blood cancer in which NK cells expressing PD-1 are known to be functionally impaired [1]. We tested ex vivo nonexpanded pNK cells taken from peripheral blood from patients with HL prior to therapy, to establish the base levels of ex vivo IRE1α-endonuclease activity. XBP1s expression was quantified by flow cytometry in HRS-stimulated pNK cells from patients with HL and compared to age-matched healthy participants. Mononuclear CD3⁻CD14⁻CD19⁻CD56⁺ pNK cells were enumerated and classified into conventional pNK cell subsets based on CD56 and CD16 expression: CD56^brightCD16^−ve and CD56^dimCD16⁺. Following HRS stimulation, XBP1s expression was reduced in both CD56^brightCD16^−ve (shown in Fig. 5a) and CD56^dimCD16⁺ subsets (shown in Fig. 5b) in HL patients, indicating that IRE1α-endonuclease activity (required to effectively process the functional XBP1s isoform) is reduced in pNK cells from cancer patients.

Next, we evaluated if there is still a remnant role for the impaired IRE1α pathway in the regulation of pNK cell effector functions in HL. As the numbers of intratumoral pNK cells in patients with HL are limited, circulating pNK cells from pre-therapy blood samples in patients with HL were expanded in vitro. Then the functional impact of IRE1α blockade (using IRE1α-inhibitor 4μ8c) was tested in the presence of HDLM2 target cells. This showed a reduction of IFNγ (shown in Fig. 5c, d) and TNFα cytokines release (shown in Fig. 5e, f), CD107a degranulation (shown in Fig. 5g, h), HRS target cell killing (shown in Fig. 5i), and F-actin accumulation at the NKIS (shown in Fig. 5j, k) in pNK cells from patients with HL.

These results strongly suggest that IRE1α pathway still retains a residual activity and thus it may be possible to restore a “healthy” homeostasis. As our previous study demonstrates that pNK cell effector functions are significantly improved by the immunotherapy medication Pembrolizumab (a monoclonal antibody designed to target and block PD-1) [1], we sought to determine whether PD-1 blockade restores NK cell effector functions that are reduced due to impairment of IRE1α pathway. Cytokine secretion (IFNγ, TNFα), degranulation (CD107a), HRS target cell killing, and F-actin accumulation at the NKIS were tested in pNK cells from patients with HL. We tested the effect of Pembrolizumab (α-PD-1) on pNK cell functions upon IRE1α inhibition (shown in Fig. 5c–k). PD-1 blockade restored both IFNγ (shown in Fig. 5c, d) and TNFα (shown in Fig. 5e, f) cytokines release in pNK cells (except for IFNγ in CD56^dimCD16⁺). However, PD-1 blockade did not restore degranulation (shown in Fig. 5g, h) and cytotoxicity (shown in Fig. 5i) functions, despite a significant increase in immune synapse formation when PD-1 is blocked (shown in Fig. 5j, k), in pNK cells from patients with HL. Moreover, we observed that pNK cells from patients treated with Pembrolizumab alone have restored IFNγ release (shown in online suppl. Fig. S4A, B) and immune synapse (shown in online suppl. Fig. S4H, I), but none of the other effector functions were enhanced (shown in online suppl. Fig. S4), suggesting that the improvement in NK cell function induced by PD-1 blockade is only partially by rescuing IRE1α-pathway impairment.

In confirmatory experiments, we also measured cytokine secretion, degranulation, and cytotoxicity upon IRE1α blockade on the SNK10 cell line stimulated with K562 and HDLM2 target cells. The results showed a significant reduction in all effector functions including IFNγ release (shown in online suppl. Fig. S5A, B), TNFα release (shown in online suppl. Fig. S5C, D), CD107a degranulation (shown in online suppl. Fig. S5E), and lysis of target cells (shown in online suppl. Fig. S5F), confirming that the IRE1α pathway has an important role in NK cell effector function. Moreover, IRE1α blockade efficiently inhibits XBP1 splicing in NK cells when exposed to all three tested target cells (K562, KMH2, and HDLM2; shown in online suppl. Fig. S5G), further confirming the IRE1α-pathway involvement in XBP1 splicing in NK cells. We also tested if the IRE1α pathway has a role in NK cell movement and synapse formation. SNK10 cell motility was assessed by live cell imaging (shown in online suppl. Fig. S5H). IRE1α-blocked NK cells traveled shorter distances and at lower velocity (shown in online suppl. Fig. S5H).

As a recent study has reported that NK cells can acquire functional PD-1 transferred from cancer cells via trogocytosis [47], we also investigated whether the PD-1 expression detected in the cellular membrane of NK cells used in our assays could have been acquired from malignant HRS cells due to trogocytosis in HL instead of an endogenous expression in NK cells. First, PD-1 expression was tested in a range of HL cell lines and none of the tested lines expressed PD-1 mRNA or protein (shown in online suppl. Fig. S6A–C). Then, we tested the expression of PD-1 and CD30 (a marker of HRS cells) on pNK cells in pre-therapy blood samples in patients with HL. The results demonstrate that pNK cells express PD-1 but not CD30 (shown in online suppl. Fig. S6D), supporting the notion that PD-1 expression on pNK cells is not due to trogocytosis but an endogenous expression in pNK cells.

Having established that the IRE1α pathway is impaired in the pNK cells of patients with HL, we next compared the native expression of miR-34a-5p and PD-1 in pNK cells from 9 healthy individuals and 11 patients with HL. Our results with NK cell lines indicate that miR-34a-5p negatively regulates PD-1 expression; therefore, an inverse relationship is expected to emerge between miR-34a-5p and PD-1 expression in pNK cells. We used a PrimeFlow RNA assay to simultaneously measure expression of miR-34a-5p and PD-1 protein by flow cytometry. With the specific oligonucleotide target probe, the assay hybridizes and amplifies miRNA molecules at single-cell resolution without the requirement for RNA extraction and quantification by conventional qRT-PCR. We observed two distinct clusters based on miR-34a-5p and PD-1 expression in pNK cells from patients with HL (cluster 1: CD56⁺CD3^−vePD-1⁺miR-34a-5p^low and cluster 2: CD56⁺CD3^−vePD-1⁺miR-34a-5p^high; shown in Fig. 6a). We then tested the two subgroups for correlation between miR-34a-5p and PD-1 expression with the CD56⁺CD3^−vePD-1⁺miR-34a-5p^high subgroup, showing a significant negative correlation (Pearson’s correlation coefficient R = −0.92; p = 0.009; shown in Fig. 6a), while the CD56⁺CD3^−vePD-1⁺miR-34a-5p^low subgroup showed a negative correlation; however, this did not reach statistical significance (Pearson’s correlation coefficient R = −0.83; p = 0.08; shown in Fig. 6a). No significant negative correlation was found in healthy subjects with detectable levels of PD-1 expression (Pearson’s correlation coefficient R = 0.18; p = 0.73; data not shown). This inverted relationship confirms our observations in NK cell lines in which miR-34a-5p is identified as a key inhibitory regulator of PD-1 expression.

Intriguingly, we also found miR-34a-5p expression is reduced in pNK cell cluster 1 (CD56⁺CD3^−vePD-1⁺miR-34a-5p^low subgroup) compared to healthy individuals but not in cluster 2 (CD56⁺CD3^−vePD-1⁺miR-34a-5p^high, shown in Fig. 6b). Arguably, the impaired IRE1α pathway observed in pNK cells from patients with HL may also impact the regulation of miR-34a-5p expression and consequently PD-1 expression. However, a more detailed understanding of the mechanistic nature of IRE1α-pathway impairment in cancer will require further studies with an appropriate design to address these specific clinical questions.

Herein we uncovered a new mechanistic association between the miRNA miR-34a-5p, transcription factor XBP1, and immune checkpoint PD-1 upon IRE1α-pathway activation to modulate human NK cell effector function and tolerance in HL. Activation of NK cells involves the IRE1α pathway, which triggers a rapid change in XBP1s-mediated gene regulation. This is amplified by an IRE1α endonuclease-mediated reduction of miR-34a-5p. Notably, miR-34a-5p inhibits XBP1 and PD-1 expression. XBP1s also directly promotes PD-1 expression that is further enhanced by the IRE1α-mediated reduction of miR-34a-5p, thus fine-tuning NK cell effector function. However, the IRE1α pathway is impaired in NK cells within patients with HL, resulting in a disruption of the XBP1/miR-34a axis and of PD-1 expression within NKcells of patients with HL. The mechanisms by which HL cells influence NK cells would be an interesting area of future research.

Several studies indicate that the NK cell/tumor cell PD-1/PD-L1 axis is involved in immune evasion in blood cancers including HL [1, 5, 48]; however, the molecular mechanisms that fine-tune PD-1 expression are unknown. The findings from the current study indicate that the IRE1α pathway regulates PD-1 in HL. Upon initial contact with HRS cells, the IRE1α pathway is induced resulting in upregulation of NK cell effector function. However, under sustained immune activation, PD-1 expression is upregulated, promoting immune tolerance (or exhaustion). Arguably, this “negative feedback” serves as a regulatory mechanism to adjust to chronic NK cell activation imposed by cancer-induced stress. The second outcome of IRE1α pathway activation is to fine-tune PD-1 expression by removing the inhibitory miR-34a-5p that in turn increases PD-1 expression (shown in Fig. 7). Chronic activation of the IRE1α pathway downregulates miR-34a-5p, which increases PD-1 expression and its inhibitory effects in NK cells. In line with these findings, miR-34a-5p and PD-1 were inversely correlated in pNK cells of patients with HL, implicating the IRE1α/miR-34a-5p/PD-1 axis as a novel mechanism in NK cell linking immune cell activation and tolerance through IRE1α and immune checkpoint PD-1.

Recent studies have demonstrated IRE1α as a “double-edged sword” that co-opts its endonuclease activity to regulate the activation of XBP1s while simultaneously repressing a subset of miRNAs [27]. IRE1α-endonuclease activation causes a rapid decay of CUGCAG motif-containing miRNAs (i.e., miR-17, miR-34a, miR-96, and miR-125b) and hence strongly increases the expression of several genes affecting multiple cellular pathways [27, 42]. The current study is the first to provide experimental evidence that IRE1α endonuclease works as a dual regulatory switch to enhance PD-1 expression in the NK cells of patients with HL, through simultaneous activation of XBP1s and repression of miR-34a-5p. Interestingly, neither PD-1 mRNA or protein was expressed across HRS cell lines, nor was the HRS marker CD30 observed in pNK cells in patients with HL, indicating that the expression of PD-1 in the NK cells of patients cannot be explained by trogocytosis as previous published [47]. The IRE1α/miR-34a/PD-1 axis may be one factor to explain some contradictory findings regarding PD-1 expression on NK cells [1, 5, 49, 50].

Notably, miR-34a-5p has been previously found to be a master regulator of tumor suppression [34], and to enhance the repression of PD-L1 in acute myeloid leukemia and lung cancer [28, 29]. Therefore, the IRE1α/miR-34a regulatory circuitry in the PD-1/PD-L1 immune checkpoint axis may partially explain the immune exhaustion commonly observed in other blood and solid cancers [34]. Notably, T-cells are abundant within the tumor microenvironment (TME), but only a few possess the appropriate T-cell receptor to recognize antigens expressed on HRS cells. In contrast, NK cells are not as abundant in the TME and they do not induce lysis through the recognition of specific antigens. However, unlike T-cells, NK cells tend to target cells with reduced or lost expression of MHC-class I. This indicates that the impairment of NK cell function is an important mechanism of immune evasion in HL. We also found that PD-1 blockade only partially reversed NK cell dysfunction and neither degranulation nor cytotoxic function was restored by PD-1 blockade after IRE1α-pathway inhibition. This indicates that alternative strategies are required to restore NK cell mediated anti-HL immunity. Hence, our findings support the current growing interest in small RNA-based therapies to manipulate immune checkpoints in immune cells including NK cells to improve cancer immune therapies in blood cancers [51].

In conclusion, a hitherto undescribed NK cell dual regulatory axis is outlined, by which the IRE1α pathway fine-tunes NK cell effector activity through an interconnected activation of the transcription factor XBP1s and inhibition of miR-34a-5p to modulate PD-1 expression. The IRE1α pathway is impaired in NK cells within patients with HL, resulting in disruption of the dual role IRE1α pathway plays in regulating the XBP1/miR-34a axis and PD-1 expression within NK cells. IRE1α endonuclease in NK cells represents a potential target to optimize NK cell immunotherapeutic strategies in future HL treatments.

The work utilized the flow cytometry core facility at the Translational Research Institute, Brisbane, QLD, Australia.

This study protocol was reviewed and approved by Metro South Health Human Research Ethics Committee (HREC/07/QPAH/035). Written informed consent was obtained for all participants in this study in accordance with Human Research Ethics Committee guidelines.

The authors declare no competing nonfinancial interests but the following competing financial interests: Maher K. Gandhi receives research funding from Janssen and Beigene and has received honoraria from Novartis.

This work was supported by the Leukaemia Foundation and the Mater Foundation (M.K.G.), Princess Alexandra Hospital Foundation (M.K.G. and A.S.C.), and University of Queensland PhD scholarships for international students (G.T. and K.B.). The Translational Research Institute is supported by the Australian Government.

A.S.C., M.K.G., S.M., J.N., K.B., G.T., and M.K.G. conceived and designed the study; M.K.G. provided study materials and/or patients; A.S.C., K.B., G.T., M.B.S., S.M., J.N., L.M.L., F.V., Q.C., S.C.L., A.P., and J.G. collected and assembled data; A.S.C., K.B., G.T., M.K.G., S.M., and J.N. provided data analysis and interpretation; A.S.C. and M.K.G. provided financial support; and all authors undertook manuscript writing and final approval and are accountable for all aspects of the work.

Karolina Bednarska and Gayathri Thillaiyampalam contributed equally to this work, and Maher K. Gandhi and Alexandre S. Cristino shared senior authorship.

All data generated or analyzed during this study are included in this article and its online supplementary material files. Further inquiries can be directed to the corresponding authors.