Enhanced antitumoral activity of TLR7 agonists via activation of human endogenous retroviruses by HD

In this work, we are reporting that “Shock and Kill”, a therapeutic approach designed to eliminate latent HIV from cell reservoirs, is extrapolatable to cancer therapy. This is based on the observation that malignant cells express a spectrum of human endogenous retroviral elements (HERVs) which can be transcriptionally boosted by HDAC inhibitors. The endoretroviral gene HERV-V2 codes for an envelope protein, which resembles syncytins. It is significantly overexpressed upon exposure to HDAC inhibitors and can be effectively targeted by simultaneous application of TLR7/8 agonists, triggering intrinsic apoptosis. We demonstrated that this synergistic cytotoxic effect was accompanied by the functional disruption of the TLR7/8-NFκB, Akt/PKB, and Ras-MEK-ERK signalling pathways. CRISPR/Cas9 ablation of TLR7 and HERV-V1/V2 curtailed apoptosis significantly, proving the pivotal role of these elements in driving cell death. The effectiveness of this new approach was confirmed in ovarian tumour xenograft studies, revealing a promising avenue for future cancer therapies.

Toll-like receptors (TLRs) play a pivotal role in innate immunity by recognising pathogen-associated molecular patterns1. TLR3, 7, 8 and 9 are located in the endosome and collectively play a critical function in the detection of viral signatures. TLR7/8 are phylogenetically and structurally very similar and both can recognise GU-rich sequences in single-stranded RNAs (ssRNAs) from viruses2. TLR7/8 are coupled to the adaptor protein MyD88 which activates downstream NF-κB driven-genes3. NF-κB plays an important role in the initial formation and progression of malignancies4. It triggers the expression of apoptosis inhibitors, growth and angiogenesis factors, cyclins, as well as metalloproteinases involved in tumour invasiveness5. TLR7/8 are also involved in a plethora of intracellular signalling cascades which culminate in the gene expression of pro-inflammatory cytokines and chemokines6. Both receptors are also widely expressed in several types of cancer and therefore, agonists of these TLRs are currently of interest as antitumoural compounds7,8,9,10,11.

Among the proteins aberrantly expressed in malignant cells, those derived from human endogenous retroviruses (HERVs) represent a vast group with a strong link to cancer cell biology, which makes them potential targets for therapy12,13. HERVs constitute a broad class of retroviral genetic elements, which were integrated into the host genome and are vertically transmitted in a Mendelian form to the progeny. While intact HERVs possess a typical 5′-LTR-gag-pol-env-LTR-3′ retroviral structure, HERVs in humans (and other apes) have lost the capability of horizontal transmission due to epigenetic repression and/or the accumulation of mutations, which eliminated their infective capacity. Yet, some HERV genes have retained intact ORFs that via complementation may create transducing particles which form bioactive quasi-species12,14,15,16,17,18. Some of these endoretroviral elements have also been linked to increased tumour cell heterogeneity19 and suggested to promote carcinogenesis through the very same function they perform in placenta physiology—fusogenicity and immunosuppression—to stimulate uncontrolled cell fusion and abrogate the anti-oncogenic cytolytic immune response20,21.

An Env-coding HERV sequence harboured by the human genome and expressed in the placenta (locus 19q13.41) is HERV-V. It is more ancient than the two human syncytins as it was acquired by the primate lineage more than 45 million years ago, and is found in both New and Old World Monkeys22,23. This provirus has undergone post-integrative duplication which resulted in the emergence of HERV-V1 (producing the C-terminally truncated protein EnvV1) and HERV-V2 (coding for the full-length EnvV2), which are ~34 kb apart and show high nucleotide identity. While behaving like a syncytin in Old World Monkeys, EnvV2 has lost its fusogenicity in hominoids through a slow, still-ongoing selective process which, however, did not affect its immunosuppressive activity23.

The sole expression of HERVs is not an adequate stimulus for recognition by TLR7/8, not even overexpression observed in cancer cells refractory to cytostatic therapy. This indicates that a minimum level of ssRNA with retroviral signatures does not exist, which would be needed by TLR7/8 to render cells susceptible to trigger apoptosis. Even when the HERV RNA pool is overexpressed through subtoxic isodoses of HDAC inhibitors (HDACis), no cell death is achieved. Therefore, the cellular concentration of HERV RNA is not sufficient to mimic an infection, and more factors may interfere in this context. In nature, this phenomenon is observed in viral infections in which cells do not adequately activate the innate immunity, i.e. TLR7/8 cells are not stimulated.

To overcome this problem, we extrapolated the concept of “shock and kill” (SaK) therapy to cancer treatments. SaK is now the dominant strategy aimed at eliminating HIV from tissue reservoirs and basically describes a two-step intervention: first, latency-reversing agents are used to reactivate latent HIV (the “shock”) and secondly, the “kill” phase of those virus-expressing cells using neutralizing monoclonal antibodies or TCAR against viral structures24,25,26.

In this context, HDACis as for example vorinostat and romidepsin were employed as “shock” agents. In ovarian cancer (OC) models, we observed a drastic, dose-dependent expression of HERV-V1/V2 envelope genes under the influence of HDACis. Their role in cancer cell biology is fully unknown, but they have been detected or overexpressed in malignant tumours and not in regular adjacent tissues. For the “kill” phase, instead of components of the humoral or cellular immunity, we involved the innate immunity by employing TLR7/8 agonists (TLR7/8as). Besides their antiviral properties, they also show a robust antitumoural activity which is not fully understood7,8,10,11. This combination acts synergistically and effectively induces cell death at sub-cytotoxic doses. Steady-state analyses of specific canonical signal transduction cascades like TLR7/8-NFκB, Akt/PKB, Ras-MEK-ERK and apoptotic pathways further revealed consistency with bidirectional interaction of the two substance classes. The effects observed in vitro were mirrored in ovarian carcinoma xenograft models, giving further credence to this strategy as a potential anticancer therapy.

In summary, our data indicate that the combination of HDACis acting as Latency Reversing Agents (LRAs) together with TLR7as may represent a novel, unprecedented therapeutic strategy to eliminate cancer cells. The fact that some drugs used in this study are FDA-approved might open up a new avenue for clinical trials to define therapeutic efficacy and possible adverse effects.

HERV expression in OC cell lines, ovarian tumours and surrounding tissues

A range of HERV elements was detected both at the RNA and protein level in a panel of OC cell lines and primary tumour cells (Fig. 1, Supplementary Tables 1 and 2). Interestingly, some HERV envelope proteins were expressed beyond basal levels in SKOV3CP carboplatin-resistant OC cell subline and in OvCa236 ascites-derived primary OC cells obtained from heavily treated patients (Fig. 1a). Expression of HERV-V1/V2 protein in particular was mainly confined to the tumour areas (Fig. 1b) and considerably less detectable in healthy tissue and the cancer-surrounding stroma (Fig. 1b).

The fact that these retroviral elements were overexpressed in chemorefractory cancer cells provided a first hint that they may also be susceptible to chemical activation by HDACis, which are known to act as LRAs for genome-integrated retroviruses like HIV27.

HDACis are potent de-repressors of HERV-V1/V2 transcription

In line with the above hypothesis, we investigated the role of HDACis in driving HERV transcription and specifically the transcription of genes, which encode for envelope proteins. For this purpose, we tentatively selected romidepsin and vorinostat, two therapeutics that are currently in clinical use and known to be potent HIV anti-latency drugs in the context of the SaK approach. Their IC50 values were determined over 72 h in different OC cell lines and primary cells isolated from OC ascites (Fig. 2a, Supplementary Fig. 1). Romidepsin and vorinostat were found to be highly cytotoxic at concentrations of ~4 nM and 1.5 µM, respectively. These values were used as reference in all subsequent biological experiments.

Using fractions of the IC50 values, we analysed the transcriptional activation of a panel of HERV elements (Supplementary Table 1) after 24 h of drug exposure. It was found that some HERV genes were effectively de-repressed (activated) by our HDACis among which HERV-V1/V2showed greater post-exposure transcriptional activity than the other HERVs analysed. This effect was consistent in almost all cancer cells, which we looked at (Supplementary Fig. 2).

Seeing that some HERVs could be transcriptionally activated by HDACi treatment, we asked the question whether this effect might be exploited in analogy to SaK to induce selective cancer cell death in two ways: by forcing the expression of HERV envelope proteins in the cell membrane and targeting them with specific antibodies or CART systems, and second, by recruiting the cell’s own innate immune response via agonization of TLR7/8 receptors, which recognise viral signatures. We therefore determined the IC50 of the TLR7/8as imiquimod and vesatolimod which are currently in clinical use (Fig. 2b, Supplementary Fig. 3a). Compared to HDACis, both agents proved less cytotoxic in cancer cells, with an IC50 of around 40 µM in each case.

Synergistic cytotoxic effects of HDACis and TLR7/8as in OC cell lines and primary tumour-derived cells

To obtain a first hint of the bidirectional cytotoxic activity of HDACis and TLR7/8as in OC cell lines and primary tumour cells, we performed an isobolographic analysis which combined the two drug classes (Fig. 2c, Supplementary Fig. 3b). Our isobolograms reveal that the ‘HDACi and TLR7/8a combination treatment’ (HTCT) always produced a synergistic cytotoxic effect, regardless of how the two substance classes were combined. More importantly, the HTCT resulted in a synergistic cytotoxic effect in SKOV3CP subline, which is highly resistant to carboplatin (Supplementary Fig. 3c).

HTCT is a potent, selective inducer of intrinsic apoptosis

Our studies on apoptotic pathways in OC cells revealed that the synergistic cytotoxic interaction between HDACis and TLR7/8as specifically triggers intrinsic apoptosis, confirmed in particular by detecting the cleavage of caspase 9. In addition, anti-apoptotic executors like Bcl-xL were downregulated at the protein level, thus favouring apoptotic processes28 (Fig. 3a, Supplementary Figs. 4 and 5).

Moreover, immunochemical analysis with ICC, FACS and WB performed in SKOV3WT and primary OC cells showed that HTCT led to extensive PARP (poly-ADP ribose polymerase) cleavage (Fig. 3a–c, Supplementary Figs. 4 and 5). Dose-dependent expression of cleaved caspases 3 and 9 as well as PARP was observed in vorinostat and imiquimod treated SKOV3WTcells (Supplementary Fig. 6a). The level of apoptosis was identical in the cell population, which remained attached to the culture surfaces and in the detached, floating populations (Supplementary Fig. 6b). Interestingly, HTCT showed no synergistic effect in triggering apoptosis of PBMCs derived from healthy individuals (Supplementary Fig. 6c).

HTCT induce an impairment on HERV-V1/V2 protein synthesis

HTCT rendered in strong synergistic effect, which was correlated with the decrease of HERV-V1/V2 proteins in OC cells, as judged by both FACS and Western blot analysis (Supplementary Fig. 7a, b). Thus, the reduced HERV-V1/V2 protein levels occasioned by the drug combinations might be coupled from cell death if we assume that the remaining attached populations are on their way to apoptosis as judged by ICC analysis (Fig. 3b). The effect of HDACis on protein synthesis is uncoupled from the upregulation of this gene at transcriptional level (Supplementary Fig. 8).

T LR7 and HERV-V1/V2 gene ablation with CRISPR/Cas9 significantly reduces apoptosis

In order to determine whether the observed pro-apoptotic effects were mediated by retroviral RNA signatures via TLR7, we separately ablated both TLR7 and HERV-V1/V2 in SKOV3WT cells using CRISPR/Cas9, without selecting for the ablated populations (Supplementary Figs. 9 and 10). Gene ablation was performed by targeting mRNA for TLR7 (MN_016562.4, Supplementary Fig. 9a) or HERV-V1 (NM_152473.2, Supplementary Fig. 10a) using specific gRNA designed with an IDT tool (https://eu.idtdna.com/site/order/designtool/index/CRISPR_CUSTOM) and checked against the whole human genome to determine on/off-targets with the Basic Local Alignment Search Tool (BLAST). The ablation effectiveness was confirmed by T7 digestion and qPCR (Supplementary Figs. 9b, c and 10b, c). In addition, the comparative protein expression for HERV-V1/V2 in wild-type and KO cells was assessed (Supplementary Fig. 10d). In the SKOV3HERV-V1/V2KO cells, the amplicons for HERV-V1/V2 were ~50% less expressed after HDACis exposure compared to the wild-type cells (Supplementary Fig. 11a).

As shown above, HTCT induced a strong apoptotic response in several cell systems. In contrast, apoptosis in SKOV3TLR7KO cells was noticeably less pronounced as determined by PARP cleavage (Fig. 4a, b), indicating that TLR7 is a pivotal mediator of post-treatment cell death. Similar to TLR7, we detected a global 50% reduction in apoptotic cells in the HERV-V1/V2 knockout cultures, suggesting that the products of these genes are indeed mediators of apoptosis (Fig. 4c–e). A microscopic investigation confirmed that the SKOV3HERV-V1/V2KOsubline was insensitive to romidepsin/vesatolimod combination treatment (Supplementary Fig. 11b).

The significant reduction in apoptotic cells after TLR7 and HERV-V1/V2 gene ablation validates the involvement of these elements in mediating cell death driven by HTCT, and suggests a possible extrapolation of the SaK concept to cancer therapy.

Do other HERV elements play into the SaK effect?

From dose-dependent studies, we observed that HERV-V1 looks more HDACis-activatable than the truncated HERV-V2 form (Fig. 5a). Although we had established that the HERV-V1/V2gene products act as TLR7 substrates, a subsequent qPCR analysis hinted that HDACi treatment de-represses further HERV elements in OC cells (Supplementary Fig. 2). In order to study this issue in detail and pinpoint the possible contribution of other HERVs, we performed a ChIP-seq in SKOV3WT cells exposed to romidepsin and vorinostat. Analysing the histone 3 (H3) acetylation patterns after the treatment, we concluded that H3AcK9, which is known to be involved in transcriptional activity, was the most suitable candidate for chromatin immunoprecipitation (Supplementary Figs. 12 and 13).

To check the acetylation levels at the positions harbouring HERV loci and assess their variation after the treatment with HDACis, the univocal genomic coordinates of about 3280 HERV loci have been compared with the position of the H3AcK9 peaks identified by ChIP-seq analysis (Supplementary Fig. 14, Supplementary Data 1). Overall, in untreated cells, around 0.7% of peaks corresponded to 1233 individual HERV loci (37.6% of the whole dataset), and the number of co-localised peaks was increased of 2.3–2.5-fold after the treatment with HDACis, mapping to a percentage of HERV loci from 48 to 60% of the total dataset (Supplementary Fig. 14). This first analysis confirmed that HDACis are able to modulate the acetylation of repetitive elements, including HERV sequences. Subsequently, we considered only the differential peaks (diffpeaks), i.e. the peaks showing significant variation after the treatment with vorinostat and/or romidepsin. Interestingly, even if more than half of HERV loci co-localised with acetylation peaks after HDACi treatment, only a minority of them (0.9–1.9%) co-localised with diffpeaks, showing mostly a reduction in their acetylation levels (Supplementary Fig. 14, Supplementary Data 1). Moreover, the exposure to different HDACis led to the variation of different subset of HERV-loci acetylation, even if a proportion of HERV sequences (21/82) were common to both treatments (Supplementary Fig. 14). Diffpeaks co-localised HERV loci could be classified into 24 HERV groups and included the ERV-V2 locus (Supplementary Fig. 14).

We focused then on the 11 HERV loci which co-localised with diffpeaks showing an increase in acetylation levels, hence indicating an increased transcriptional activation after HDACi treatment (Table 1). Remarkably, diffpeaks corresponding to the ERV-V2 locus were identified in all samples and were the most significantly upregulated in both vorinostat and romidepsin treatment (Table 1, Supplementary Fig. 14). This was in part confirmed by ChIP-qPCR in different OC cells (Supplementary Fig. 15, Supplementary Table 3). Beside ERV-V2, only the HML8 sequence at locus 2q11.2 showed increased acetylation levels in all treated samples, independent from the HDACis used, but having lower statistical significance (Table 1, Supplementary Data 1). Given that ERV-V2 locus showed a specific increase of acetylation levels following HDACi stimulation, we assessed the position of diffpeaks within its retroviral structure and compared it with respect to the corresponding peak in untreated control (Fig. 5b). The analysis revealed that the treatment with HDACis did not only account for the increase of ERV-V2 acetylation, but also led to the shift and concentration of acetylation peaks at the 5′ region of the HERV provirus. This observation allows us to hypothesize that HDACi treatment could not only increase ERV-V2 transcriptional activation, but also modify its pattern of expression, possibly favouring HERV-specific promoter activity from ERV-V2 LTR or an upstream solitary LTR43 element, and eventually leading to alternative splicing (Fig. 5b). This scenario is also supported by the presence of promoter and proximal enhancer-like signatures that are collinear with HDACi-associated diffpeaks.

Studies on TRL7 signalling validate the mechanism of SaK in cancer cells

TLR7 transduces signals by recruiting the MyD88 adaptor protein, which in turn communicates with the transcription factor (TF) NFκB via a sequence of transducers. NFκB helps initiate the transcription of a plethora of genes implied in the inflammation network, cell survival and differentiation, and it is also found upregulated in cancers29. We observed that HTCT led to the functional disruption of this pathway (Fig. 6a). We incubated different carcinoma cell lines with both drug classes simultaneously (0.5× IC50 for 24 h) and studied how the RNA and protein levels of pivotal transducers were affected. The transcription of TLR7, as the first player in this pathway, was not negatively affected by HTCT (Fig. 6b). In contrast, protein synthesis of MyD88 was selectively impaired, suggesting a disruption of the downstream signalling switch of this transducer (Fig. 6a, c, Supplementary Table 4).

Given that SARM1 has been proposed to be an adaptor protein for TLR730, we investigated its role in TLR7 downstream signalling and in mediating apoptosis in response to HTCT. To address this question, we ablated SARM1 using CRISPR/Cas9 in SKOV3WT cells, incubated the cells simultaneously with romidepsin and vesatolimod (1× IC50) and checked for apoptosis using c-PARP as a sentinel for cell death. Although SARM1 protein tended to be reduced after HTCT, the difference in apoptosis between ablated and normal cells was not significant (Supplementary Fig. 16), indicating that SARM1 is not likely to play a role as a TLR7 adaptor in this cell line.

Following romidepsin/vesatolimod combination treatment, the protein synthesis of undigested NFκB1 (p105 form) was reduced, suggesting that I-κBα was hypophosphorylated, the NFκB/I-κBα complex did not dissociate, and NFκB1 was not transcribed; this effect was confirmed a posteriori with immunotechniques (Fig. 6a). It means that, after treatment, the full-length form of the canonical transcription factor NFκB1 cannot translocate into the nucleus and perform its role as TF, which was corroborated with a luciferase system containing a consensus sequence for this TF coupled to a firefly luciferase as reporter (Fig. 6d). Our experiment established that NFκB is not functional even if stimulated with TNFα and hence cannot initiate the transcription of genes which are in part governed by it, including the gene for NFκB itself. A critical NFκB target gene is Bcl-xL31, which was selectively reduced at the protein level after HTCT (Fig. 3), confirming the disruption of this transcription factor. IL8 appeared to be less affected, as its transcription was downregulated under the influence of either HTCT in OvaCar3 and OvCa2810, but less so in SKOV3WT. What we observed in OvarCa3 and OvCa2810 is a compensatory homoeostatic behaviour of the axis: IL downregulation vs. CXCR1 upregulation (Supplementary Fig. 17). IL8 is a rather promiscuous interleukin, owing to its ability to dock several chemokine receptors, the most prominent being CXCR1 followed by CXCR2. IL8 is upregulated in several cancer types, including OC, and is subject to transcriptional regulation by NFκB and other TFs32.

We also examined the impact of HTCT on inflammatory signalling at the RNA level in SKOV3WT, which are also dependent on NFκB. As depicted in Fig. 6e, the expression profile of some inflammatory mediators was very similar for the two combinations but differed when the drugs were applied individually (Supplementary Table 4). IFNɣ and TNFα, in contrast, were substantially upregulated in all cases. These inflammatory factors were also analysed partially ex vivo using PBMCs from healthy individuals (Supplementary Fig. 18).

HTCT disrupts other signal transduction pathways

In an effort to further clarify the mechanism(s) leading up to the apoptotic response upon HTCT, we investigated canonical signal transduction pathways such as Akt/PKB and Ras-MEK-ERK in different OC cell lines. These pathways are tightly interconnected with NFκB signalling, as I-κBα, for example, is known to be phosphorylated via Akt/PKB33,34.

Western blot analysis of specific transducers revealed that all drug combinations reduced the survival protein Akt at the protein level and suppressed its phosphorylated form entirely, while imiquimod alone induced its hyperphosphorylation (Fig. 7a, Supplementary Figs. 19 and 20). The dephosphorylated and phosphorylated forms of β-catenin and c-Myc were downregulated at the protein level, similar to Akt. Taken together, Akt activation was fully disrupted by all analysed HTCT.

As previous works have reported that some HERVs are associated with the activation of Ras-MEK-ERK35 cascade, we investigated the effect of HTCT on this pathway. In fact, a disruption was seen in this cascade. While ERK1/2 and MEK1/2 protein levels were not reduced after treatment with either HDACis, their phosphorylated forms were depleted, while the opposite effect was observed after treatment with the TLR7 agonists. In cancer cells, MyD88 constitutively inhibits MKP3 phosphatase activity on MEK1/2, which results in constitutive MEK1/2 activation. We showed that MKP3 protein expression was not affected by any of the analysed drug interventions. Since MyD88 was downregulated by each of the HTCTs, we expected MEK1/2 to be dephosphorylated. Yet, the dephosphorylation of its downstream transducer under HDACi treatment suggested that other mechanisms might be at play that go beyond the interaction between MyD88 and MKPs. Figure 7b depicts global disruption points of the TLR7-NFκB, Akt/PKB, and Ras-MEK-ERK cascades upon HTCT in OC cells.

Synergistic anticancer activity in an SKOV3WT xenograft mouse model

The cell lines SKOV3WT and OvCar3WT generate well-differentiated adenocarcinomas in xenograft nu/nu mice. We studied the impact of romidepsin and vesatolimod administered either individually or in combination on tumour growth and metastasis formation in an SKOV3WT xenograft mouse model (Fig. 8). A marked antitumoral effect was observed after four rounds of treatment. While romidepsin and vesatolimod produced an antitumoral effect when administered alone, their combination mirrored the synergistic effects seen in our in vitro experiments (Fig. 8a). The combination also showed an effect on the colonisation of SKOV3WTcells in lungs (Fig. 8b) which mirrored the antitumoral effect observed when tumour cells were implanted subcutaneously.

We also looked at the impact of the HTCT on HERV-V2 mRNA and protein levels by treating OvCar3 cells with lower drug doses compared to SKOV3WT and subjecting the remaining tumour mass to qPCR and IHC: HERV-V2 mRNA and protein levels mirrored the observed in vitro effects after romidepsin-vesatolimod combination treatment (Fig. 8c), while vesatolimod alone had a distinct reducing effect (Fig. 8d).