NRF3 suppresses squamous carcinogenesis, involving the unfolded protein response regulator HSPA5

Abstract Epithelial skin cancers are extremely common, but the mechanisms underlying their malignant progression are still poorly defined. Here, we identify the NRF3 transcription factor as a tumor suppressor in the skin. NRF3 protein expression is strongly downregulated or even absent in invasively growing cancer cells of patients with basal and squamous cell carcinomas (BCC and SCC). NRF3 deficiency promoted malignant conversion of chemically induced skin tumors in immunocompetent mice, clonogenic growth and migration of human SCC cells, their invasiveness in 3D cultures, and xenograft tumor formation. Mechanistically, the tumor‐suppressive effect of NRF3 involves HSPA5, a key regulator of the unfolded protein response, which we identified as a potential NRF3 interactor. HSPA5 levels increased in the absence of NRF3, thereby promoting cancer cell survival and migration. Pharmacological inhibition or knock‐down of HSPA5 rescued the malignant features of NRF3‐deficient SCC cells in vitro and in preclinical mouse models. Together with the strong expression of HSPA5 in NRF3‐deficient cancer cells of SCC patients, these results suggest HSPA5 inhibition as a treatment strategy for these malignancies in stratified cancer patients.


Introduction
Skin cancers are the most frequently diagnosed malignancies in humans, with one in three cancers diagnosed as a malignant skin tumor.The two major types of epithelial skin cancers, basal and squamous cell carcinomas (BCC and SCC), are diagnosed in 2-3 million people world-wide every year (Cia z _ zy nska et al, 2021; Sung et al, 2021).They develop in a complex multistage process, which ends up with cells that have acquired an altered proliferative capacity, invasiveness, and possibly metastatic potential (Rundhaug & Fischer, 2010).SCCs are often preceded by precursor lesions, of which actinic keratosis (AK) is the most frequent one that develops at multiple sites, in particular in chronically sun-exposed skin of aged individuals (Werner et al, 2013).
The development and progression of epithelial skin cancers is controlled by various transcriptional regulators, which orchestrate the expression of downstream targets that affect proliferation, survival, and/or migration of tumor cells.A key player in cancer development is nuclear factor (erythroid-derived 2)-like 2 (NFE2L2; NRF2; Sporn & Liby, 2012;Rojo de la Vega et al, 2018).Loss of Nrf2 predisposes mice to different types of tumors, including skin SCCs, because of its important role in the detoxification of reactive oxygen species (ROS) and toxic chemicals (auf dem Keller et al, 2006;Xu et al, 2006;Sporn & Liby, 2012;Schmidlin et al, 2021).On the other hand, NRF2 is frequently hyperactivated in cancer cells, including SCC cells (Cescon et al, 2015;Oshimori et al, 2015), which promotes their malignancy and chemoÀ/radioresistance (Sporn & Liby, 2012;Schmidlin et al, 2021).
In contrast to the well-characterized NRF2, little is known about the target genes, mechanisms of action, and roles in cancer development and progression of the related NRF3 (NFE2L3) protein.NRF3 mRNA levels are strongly elevated in various types of cancers, and NRF3 was identified as one of the most significantly mutated genes across 12 cancer types (Kandoth et al, 2013).The few published functional studies suggest a pro-tumorigenic function of NRF3 in most tissues, in particular in colorectal cancer, by controlling cancer cell growth and transcriptional regulation of proteasome genes (Aono et al, 2019;Bury et al, 2019;Waku et al, 2020).However, positive or negative correlation of high NRF3 mRNA levels with patient survival was described depending on the type of cancer (see https://www.proteinatlas.org/ENSG00000050344-NFE2L3/pathology).
In the skin, NRF3 protein is highly expressed in basal keratinocytes (Braun et al, 2006;Siegenthaler et al, 2018).We previously showed that its loss protects keratinocytes in vitro and in vivo from apoptosis induced by UVB and other insults in an NRF2-independent manner (Siegenthaler et al, 2018), suggesting that the high levels of NRF3 in basal keratinocytes promote the elimination of mutated/damaged cells.Hence, we speculated about a possible role of NRF3 in skin carcinogenesis, as UVB radiation is the primary cause for skin cancer (Zambrano-Rom an et al, 2022).
Here we show that NRF3 protein expression is downregulated in invasively growing skin cancer cells of BCCs and SCCs in humans, suggesting a tumor-suppressive function in the skin.This was confirmed in functional 2D and 3D culture studies as well as in xenograft and chemically induced skin cancer models.Surprisingly, this activity is not mediated at the level of transcription, but results at least in part from NRF3-mediated destabilization of heat shock protein family A (HSP70 Member 5; HSPA5), a key regulator of the unfolded protein response (UPR) in the endoplasmic reticulum (ER).Upon loss of NRF3, HSPA5 levels increased, thereby promoting cancer cell survival and migration.These findings offer promising therapeutic opportunities for the treatment of epithelial skin tumors with HSPA5 inhibitors as demonstrated in this study using preclinical mouse models.

NRF3 protein expression is downregulated in human non-melanoma skin cancer (NMSC)
To determine a potential role of NRF3 in NMSC, we analyzed its expression in BCCs and SCCs and in AK precursor lesions.NRF3 mRNA levels in whole skin/tumor samples did not differ between AK, BCC, and normal human skin, while an increase was seen in 3 out of 9 SCCs (Fig 1A).
A strong perinuclear staining of NRF3 was observed in basal keratinocytes of normal human epidermis and in the outer root sheath of hair follicles, while NRF3 was not or only hardly detectable in the dermis (Fig 1B).In contrast to the mRNA data, NRF3 staining was reduced in AK lesions, and a strong downregulation or even complete loss of NRF3 was observed in the invasively growing tumor cells of BCCs and SCCs (Fig 1B and C).This was confirmed by analysis of data presented in the Human Protein Atlas, which revealed that 8 out of 12 epithelial skin cancers had lost NRF3 in the tumor cells.Three tumors showed weak staining for NRF3, and only one tumor had medium expression similar to keratinocytes of normal skin (Fig 1D; https://www.proteinatlas.org/ENSG00000050344-NFE2L3/pathology/skin+cancer).
Together, these results identify NRF3 downregulation at the protein level as a hallmark of invasively growing skin cancer cells in vivo.Consistently, analysis of publicly available data revealed that amplification of the NRF3 gene, which is frequent in other tumors, does not occur in NMSCs.Rather, they exclusively show mutations, which were, however, not further defined (Fig 1E).

Loss of Nrf3 promotes growth and malignant conversion of chemically induced skin tumors
To determine if the loss of NRF3 in skin cancers is functionally relevant, we performed a two-stage skin carcinogenesis study in mice with a global Nrf3 knockout (Nrf3-ko mice), which do not show phenotypic abnormalities under homeostatic conditions (Derjuga et al, 2004).Tumors were induced by a single topical treatment with the mutagen 7,12-dimethylbenzo(a)anthracene (DMBA), followed by weekly topical treatment with the tumor promoter 12-O-tetradecanoylphorbol-13-acetate (TPA).Fifty percent of the mice were sacrificed at the end of the 26-week TPA treatment period, while the other 50% were only sacrificed 20 weeks later, allowing time for tumor progression (Abel et al, 2009).Although the overall number of tumors per mouse was low due to the general resistance of C57BL/6 mice to skin tumorigenesis (DiGiovanni et al, 1993), Nrf3ko mice developed tumors earlier after DMBA treatment compared to their wild-type (wt) controls (Fig 2A -C).There was no more difference in tumor incidence and multiplicity after week 24 (Fig 2B and C).We noticed, however, that Nrf3-ko mice developed more mid-and in particular large-size tumors, while wt mice had mainly small tumors.This was confirmed by histomorphological analysis (Fig 2D).The larger tumor size was at least in part the result of increased keratinocyte proliferation in the papillomas (Fig 2E).
Most of the tumors in mice of both genotypes were histologically classified as acanthopapillomas.Importantly, however, the malignant conversion rate was strongly and significantly increased at week 30 and week 46 after DMBA treatment in Nrf3-ko compared to wt mice (Fig 2F).Almost 20% of the tumors in Nrf3-deficient mice were classified as malignant SCCs at week 30 and 30% at week 46, while the malignant conversion rate in wt mice did not exceed 5%.
There were no alterations in the number of dermal and epidermal T cells (CD3 + ), neutrophils (Ly6G + ), and mast cells (toluidine blue [TB + ]; Fig EV1A -D), but functional differences in immune cells cannot be excluded.Although the experiment was performed in mice with a global Nrf3 knockout, the highly predominant expression of NRF3 in keratinocytes of human skin (Fig 1B) and of mouse skin (Siegenthaler et al, 2018) and the low or even undetectable expression of Nrf3/Nfe2l3 in different types of immune cells of healthy mice (http://rstats.immgen.org/Skyline/skyline.html) suggest that Nrf3 acts mainly in a cell autonomous manner in keratinocytes during tumor progression.Therefore, we next analyzed the functional consequences of NRF3 deficiency in SCC cells.

Loss of NRF3 promotes malignancy of SCC13 cells in vitro and in 3D organotypic skin cultures
To study the consequences of NRF3 deficiency in human SCC cells, we performed CRISPR/Cas9-mediated knockout of NRF3 in the weakly malignant human SCC13 cell line (Rheinwald & Beckett, 1981).These cells express NRF3, which was stabilized by the proteasome inhibitor .Following clonal expansion, the knockout was confirmed by Western blot analysis (Fig 3A).SCC13 cells, which had been transduced with empty vector (EV) lacking a single guide (sg)RNA and which were also clonally expanded, were used as controls.There was no compensatory increase in NRF1 or NRF2 mRNA or protein levels as shown by qRT-PCR (Appendix Fig S1A) or Western blot analysis of lysates ).However, it strongly promoted clonogenicity and migration of the cancer cells, whereas their proliferation was not significantly affected (Fig 3B -D).We also generated SCC13 NRF3-KO cells with doxycycline (Dox)-inducible reexpression of NRF3 (designated RescueNRF3 cells).The recombinant NRF3 protein was efficiently expressed in these cells when 20 ng/ml Dox or more were used (Fig 3E).Under these conditions, the small variant (C-form) was detectable in the nucleus, even without MG-132 treatment (Fig 3F).Similar to the endogenous NRF3, three isoforms of the recombinant NRF3 were detected, including the N-glycosylated A variant, the ER-resident B variant, and the smallest (C) variant, which is present in the nucleus (Nouhi et al, 2007;Figs 3E and EV2A).The high cell migration rate was reduced by Dox-inducible expression of recombinant NRF3 (Fig 3G).When grown in hanging drops, the NRF3-KO cells developed much larger spheroids (Fig 3H).Vice versa, Dox-inducible overexpression of NRF3 in wt SCC13 cells reduced the migration rate and spheroid size (Fig EV2A -E).The latter was also reduced by Dox-inducible re-expression of recombinant NRF3 in the NRF3-KO cells (Fig 3I).We also generated skin equivalents in which keratinocytes are seeded on top of primary human foreskin fibroblasts, which had previously deposited their own extracellular matrix (Berning et al, 2015).NRF3-deficient SCC13 cells invaded into the dermal equivalent, while this was not observed with the parental cells (Fig 3J and K).
The positive effect of loss of NRF3 on cell migration and spheroid formation without alterations in NRF1 and NRF2 expression and activation was reproduced with the malignantly transformed HaCaT cell derivative, HaCaT-Ras (Ryle et al, 1989;Fig EV3A-F).

Loss of NRF3 enhances tumorigenesis and invasiveness in a xenograft mouse model
We

NRF3 mainly acts in the endoplasmic reticulum of keratinocytes
To identify genes that are controlled by NRF3 in skin cancer cells, we performed RNA sequencing (RNA-seq) of RescueNRF3 cells, introduced in Fig 3 .Cells were treated with DMSO or Dox for 10 h to allow expression of recombinant NRF3 and translocation of the shorter C variant, which lacks the ER targeting domain (Nouhi et al, 2007), into the nucleus (Fig 3F).Surprisingly, very few differentially expressed genes with a P-value of at least 0.05 and log 2 fold change of at least 0.5 were identified upon re-expression of NRF3 in the NRF3-KO cells (Fig 5A ), and the FPKM (fragments per kilobase per million mapped fragments) values were extremely low.The only exception was NRF3 (NFE2L3) itself, demonstrating the validity of the data.We were also not able to validate the few candidate genes, including glutathione S-transferase A4 (GSTA4) and proline rich 7, synaptic (PRR7), by qRT-PCR (Fig 5B).These findings suggest that NRF3 does not have a major role as a transcription factor in SCC cells under these experimental conditions.Consistently, RNA-seq of isolated epidermis from Nrf3-ko mice and wt littermates also identified very few genes that were differentially expressed-both in untreated mice and at 24 h after irradiation with 100 mJ/cm 2 UVB (Fig EV4A and B), when the UV-protective effect of Nrf3 deficiency was clearly visible (Siegenthaler et al, 2018).
Because of the minimal effect of NRF3 on gene transcription in keratinocytes and the predominant localization of NRF3 outside the nucleus in basal keratinocytes in vivo (Fig 1B ), we speculated that NRF3 may affect skin tumorigenesis by different mechanisms.Therefore, we searched for putative NRF3-binding partners using the proximity-dependent biotin identification (BioID) technique, which allows the labeling of proteins in close proximity to the protein of interest (Hesketh et al, 2017).For this purpose, we generated HaCaT and SCC13 cell lines with Dox-inducible expression of ◀ Figure 1.Downregulation of NRF3 protein in epithelial human skin cancer.
A qRT-PCR for NRF3 relative to RPL27 using RNA from normal human skin (HS), Actinic keratosis (AK), basal cell carcinoma (BCC), and squamous cell carcinoma (SCC).N = 5 (NS and AK), N = 6 (BCC), and N = 9 (SCC) biopsies from different individuals.Mean expression in HS was set to 1. B Representative NRF3 immunofluorescence stainings of sections from HS, AK, BCC, or SCC using antibodies against NRF3 (red) and E-Cadherin (green).Nuclei were counterstained with Hoechst (blue).Scale bar: 500 lm.HSPA5 levels are enhanced in the NRF3-KO cells and in invasively growing skin cancer cells The close proximity of NRF3 with proteins involved in ER stress and the UPR pointed to a functional role of NRF3 in this pathway.To test this possibility, we treated wt and NRF3-KO cells with the ER stress inducer tunicamycin.Loss of NRF3 strongly protected from tunicamycin-induced apoptosis as shown by cleaved caspase-3 staining (Fig 6A).This is consistent with the protection of Nrf3deficient normal mouse keratinocytes from apoptosis induced by UVB irradiation (Siegenthaler et al, 2018), which also induces ER stress and activates the UPR in keratinocytes (Farrukh et al, 2014;Seo et al, 2020).The protection of NRF3-deficient SCC13 cells from tunicamycin-induced apoptosis was associated with a less pronounced increase in the apoptotic markers CHOP and cleaved PARP (Fig 6B).In control cells, the amounts of NRF3 increased in response to tunicamycin treatment.Concomitantly, processing of NRF1 occurred, but there was no difference in the amounts of NRF1 or its processing between control and NRF3-KO cells (Fig 6B).This finding suggests that the anti-apoptotic effect of NRF3 deficiency in SCC cells is not mediated by alterations of the related NRF1 protein.Remarkably, however, NRF3-KO cells showed higher levels of HSPA5, also called glucose-regulated protein 78 (GRP78) or binding immunoglobulin protein (BiP; Fig 6B)-one of the proteins found in the BioID screen with the NRF3-HA fusion protein.We focused the further functional analysis on this protein, because (i) it is the master regulator of the UPR, which is activated in many different cancers (Wang et al, 2009), (ii) it was found in the BioID screen with an SP of 1, (iii) its close proximity with NRF3 was verified for the endogenous proteins (Fig 5H ); and (iv) overexpression of HSPA5 promoted malignancy in other types of cancer (Wang et al, 2009;Wang & Kaufman, 2014).Loss of NRF3 seems to stabilize HSPA5, because HSPA5 protein levels were also higher in NRF3-KO cells under homeostatic conditions, while HSPA5 mRNA levels were not significantly affected (Fig 6C).As expected, Dox-induced expression of NRF3 in the RescueNRF3 cells reduced the amounts of HSPA5, but a strong NRF3 overexpression had no additional effect (Fig EV5A).
HSPA5 levels were also increased in the 3D organotypic cultures generated with NRF3-KO cells, especially at the invasive front (Fig 6D, areas indicated with white arrows), as well as in the nonhyperkeratotic epithelial area of DMBA/TPA-induced tumors of Nrf3-ko vs. wt mice (Appendix Fig S4).Most importantly, HSPA5 was strongly expressed in human BCCs and SCCs, in particular in ◀ Figure 2. Loss of Nrf3 in keratinocytes promotes malignant progression of chemically induced skin tumors.
Nrf3-ko and wild-type (wt) mice were treated once topically with the mutagen DMBA, followed by weekly treatment with TPA for 25 weeks.These results point to an important role of HSPA5 in the tumorsuppressing effect of NRF3 in the skin.

HSPA5 inhibition or knock-down reverts the high malignancy of NRF3-deficient cancer cells
To determine if the increased HSPA5 levels affect the tumorigenic features of SCC13 cells, we generated hanging drop spheroids in the presence of the HSPA5 inhibitors HA15 (Cerezo et al, 2016) or YUM70 (Samanta et al, 2021) or vehicle.Both inhibitors significantly reduced the spheroid size of NRF3-KO cells, while they had no or only a minimal effect on control cells (Fig 7A and B).This is consistent with findings obtained with melanoma cells, where HA15 was more effective when the concentration of HSPA5 was high (Cerezo et al, 2016).None of the inhibitors caused significant cell death at the concentration used for the spheroid assay, and there were no or only slight differences in survival between inhibitortreated control and NRF3-KO cells (Appendix Fig S5A and B).
We also performed siRNA-mediated HSPA5 knock-down in control and NRF3-deficient SCC13 (Appendix Fig S6A and B) and HaCaT-Ras cells (Appendix Fig S6C and D).A 70-80% reduction of HSPA5 levels strongly reduced the spheroid size of NRF3-KO SCC13 cells, while it had no or only a slight effect on the control cells, which had been transduced with scrambled siRNA (Fig 7C).The enhanced migration of NRF3-KO SCC13 cells was also rescued by HSPA5 knock-down (Fig 7D).The effects of HSPA5 inhibition or knock-down were verified in HaCaT-Ras cells, where we also detected higher HSPA5 protein, but not mRNA levels in the NRF3-KO cells (Fig EV5B -F).
Finally, we performed a preclinical study and determined if treatment with HA15 affects tumor growth in the xenograft model (Fig 7E).HaCaT-Ras cells were used for this purpose because they form tumors more rapidly and efficiently compared to SCC13 cells.Indeed, injection of HA15 into growing tumors attenuated the growth of tumors formed by NRF3-KO HaCaT-Ras cells (Fig 7F ), while it had no significant effect on the tumors formed by HaCaT-Ras control cells (Fig 7G).Three-way ANOVA analysis of the combined data from control and NRF3-KO cells confirmed that the effect of HA15 treatment differed significantly between genotypes (Fig EV5G ; relevant parameter shown in bold).While there was a trend for the vehicle-treated NRF3-KO tumors to be larger than the vehicle-treated EV tumors during the treatment period, this difference was not significant, which is consistent with the data shown in Fig EV3G, where the difference between EV and NRF3-KO tumors only became significant after day 30.Ki67 staining revealed reduced proliferation in the HA15-treated tumors generated by NRF3-KO HaCaT-Ras cells, whereas HA15 even promoted proliferation in the tumors generated by HaCaT-Ras control cells (Fig 7H).Furthermore, it promoted apoptosis in the tumors formed by NRF3-KO HaCaT-Ras cells, but not in the tumors formed by control cells (Fig 7I).This difference demonstrates that HSPA5 inhibition or knock-down is mainly beneficial in NRF3-deficient cells.
In summary, pharmacological inhibition or knock-down of HSPA5 reverted the enhanced malignancy of NRF3-KO cells, demonstrating that the tumor-suppressive function of NRF3 depends on its binding partner HSPA5 and the reduction in the levels of this protumorigenic UPR protein.

Discussion
We discovered a tumor-suppressive role of NRF3 in the skin, which involves its unexpected function as a regulator of the UPR.This is surprising, since several studies showed a pro-tumorigenic function of NRF3, which is consistent with the increased NRF3 mRNA levels in many types of human cancer.For example, NRF3 overexpression promoted tumorigenesis in the colon and in other tissues through enhancement of cell proliferation and proteasomal inactivation of tumor suppressors (Bury et al, 2019;Kobayashi, 2020).On the other hand, loss of Nrf3 predisposed mice to T cell lymphoblastic ◀ Figure 3. Loss of NRF3 promotes malignancy of SCC13 cells.
A Western blot of lysates of transduced SCC13 cells (two clones of empty-vector [EV]-transduced cells [EV C1; EV C2] and of NRF3-KO cells [KO C3; KO C4]), which had been treated with the proteasome inhibitor MG-132 or vehicle, using antibodies against NRF3, NRF1, NRF2, and tubulin (loading control).B Clonogenicity of SCC13 EV and NRF3-KO cells.N = 6 cultures per cell line.Left: Quantification of the area covered by colonies relative to the whole area of the well.
Right: Representative pictures of colonies after 7 days.C BrdU incorporation into SCC13 EV and NRF3-KO cells.N = 6.D Left: Quantification of the area of open scratch at 9 h post scratch wounding in percentage of the original scratch area (left).Right: Representative pictures of scratch-wounded SCC13 EV and NRF3-KO cells at 0 or 9 h (right).N = 6.E Western blot of total lysates from RescueNRF3 cells, cultured in the presence of vehicle or different concentrations of Dox to induce NRF3 expression, using antibodies against NRF3 and GAPDH (loading control).A 6 h treatment with 10 lM MG-132 was performed to increase the signal intensity.F Western blot of nuclear lysates of SCC13 RescueNRF3 cells treated with vehicle or 20 ng/ml Dox for 6, 9, or 24 h, using antibodies against NRF3 and the nuclear marker lymphoma induced by benzo[a]pyrene (Chevillard et al, 2011) and suppressed breast cancer cell metastasis and proliferation (Sun et al, 2019).Hence, the function of NRF3 in cancer development and progression may depend on the tissue and possibly the tumor stage and the oncogenic stimulus.We found a strong downregulation of NRF3 protein in NMSCs, while NRF3 mRNA levels were upregulated in some of the SCC samples.A decrease in NRF3 protein levels was also observed during the transition from benign melanocytic lesions to primary melanoma, while the mRNA levels concomitantly increased (Immonen et al, 2022).However, it is not clear if the transcripts detected in these tumors encode the wt or mutant proteins, as NRF3 is one of 127 significantly mutated genes among 12 cancer types (Kandoth et al, 2013).In the future, it will be interesting to determine if the mutations in NRF3 that occur in NMSC (see Fig 1E) result in reduced stability of the NRF3 protein.It is also possible that NRF3 mRNA is insufficiently translated in cancer cells or that the protein gets destabilized during tumor progression, e.g. as a consequence of hypoxiainduced ubiquitination and proteasomal degradation as previously shown for other tumor suppressor proteins, e.g."deleted in breast cancer 1" (DBC1; Liu et al, 2022).
Independent of the mechanism underlying the downregulation of NRF3 protein, our results demonstrate that loss of NRF3 in SCC cells is functionally important and strongly increases their malignant features in vitro and in xenograft and chemically induced skin cancer models in vivo.However, we did not observe significant differences NRF3 has a known function as transcription factor and was reported to regulate some of the NRF2 target genes, but also genes that are not controlled by NRF2 (Pepe et al, 2010;Liu et al, 2019;Ibrahim et al, 2020;Waku et al, 2021;Saliba et al, 2022).Surprisingly, however, we could not identify significantly regulated genes in NRF3-deficient SCC13 cells after inducible expression of recombinant NRF3 or in the epidermis of UVB-irradiated wild-type vs.Nrf3 knockout mice, although NRF3 was present in the nucleus under both experimental conditions.Although we cannot exclude that NRF3 regulates transcription under other experimental conditions or at other time points after its re-expression in NRF3-KO cells, these findings suggest that NRF3-mediated gene expression is suppressed in keratinocytes through yet unknown mechanisms.They also point to another mechanism of action of NRF3 in this cell type, which does not involve alterations in the expression or activity of NRF2 or NRF1.Indeed, we found that NRF3 is in close proximity and may even directly interact with many proteins involved in the UPR-both in non-tumorigenic keratinocytes and in SCC cells.The UPR is an adaptive signaling network, which consists of three main signal transducers that are activated upon accumulation of unfolded proteins, thereby restoring protein homeostasis (Hetz, 2012).However, if the ER stress cannot be resolved and if homeostasis is not restored, damaged cells undergo UPR-induced apoptosis (Fribley et al, 2009).Cancer cells frequently experience ER stress because of the harsh conditions in the tumor, and hence the UPR is often hijacked to enhance the protein folding capacity.This favors tumor cell survival and tumor growth and progression (Wang et al, 2009;Urra et al, 2016;Madden et al, 2019).It also involves inactivation of proapoptotic UPR pathways, often as a result of mutations and/or enhanced expression of ER chaperones, like HSPA5 (Wang et al, 2009).HSPA5 senses ER stress and thereby dissociates from the UPR sensors, leading to UPR activation.HSPA5 directly affects cell survival by interacting with apoptotic pathway intermediates and by blocking caspase activation (Wang et al, 2009;Madden et al, 2019).It is overexpressed in many tumor types, and its expression correlates with cancer cell survival and malignancy (Wang et al, 2009;Wang & Kaufman, 2014).It was shown to regulate proliferation, invasion, and metastasis in head and neck SCC (Cole et al, 2019) and was crucial for mutant Kras-driven lung tumorigenesis (Rangel et al, 2021).We identified HSPA5 as a putative interaction partner of NRF3.While this may simply reflect the assistance of HSPA5 in the folding of NRF3, our follow-up data revealed a functional role of HSPA5 in the tumor-suppressive effect of NRF3 in keratinocytes, and HSPA5 protein levels increased upon loss of NRF3.HSPA5 is normally retained in the ER by its KDEL motif, but ◀ Figure 5. NRF3 binds to proteins involved in the Unfolded Protein Response.
A Volcano plot of RNA-Seq data showing significantly up-(red) and downregulated (blue) transcripts in RescueNRF3 cells treated for 10 h with 20 ng/ml Dox vs. DMSO.
A P-value < 0.05 (dotted black line at Àlog 10 (P-value) = 2) and a fold change > 0.5 (dotted black line at log 2 (FC) = AE 0.5) were used as a cutoff.B qRT-PCR using RNA from SCC13 RescueNRF3 cells treated for 10 h with 20 ng/ml Dox or vehicle (DMSO) for NRF3, GSTA4, and PRR7 relative to RPL27.when its expression levels exceed a certain threshold, it can "escape" to the cell surface, in particular in stressed cells (Li & Lee, 2006).The presence of HSPA5 at the cell surface is therefore a hallmark of some cancer cells, including melanoma cells, and cell surface HSPA5 regulates various oncogenic processes, including stemness, proliferation, migration, angiogenesis, and invasiveness (Hernandez & Cohen, 2022).In NRF3-deficient SCC cells, HSPA5 levels were strongly increased, and our preliminary data also point to a higher abundance of this protein at the plasma membrane.We do not know if the regulation of HSPA5 by NRF3 involves the help of further UPR proteins that we identified in the BioID screen.Independent of this open question, NRF3 seems to play a role in the proapoptotic arm of the UPR as suggested by the reduced apoptosis of NRF3-deficient SCC cells upon ER-stress induced by tunicamycin (this study) and of UVB-irradiated murine keratinocytes in vitro and in vivo (Siegenthaler et al, 2018).In the future, it will be important to determine if NRF3 and HSPA5 directly bind to each other, how the interaction of NRF3 with HSPA5 regulates the amount and localization of this ER chaperone, and if additional mechanisms contribute to the tumor-suppressive effect of NRF3 in keratinocytes.
Independent of these open questions, the findings obtained in this study are of potential therapeutic relevance because pharmacological inhibition or knock-down of HSPA5 rescued the malignant NRF3-KO phenotype.The HSPA5 inhibitor HA15 has previously been shown to inhibit melanoma development in mice by enhancing cancer cell death, in particular in cells with high HSPA5 levels (Cerezo et al, 2016).Furthermore, inhibition of HSPA5 by HA15 induced apoptosis in lung cancer cells (Wu et al, 2020).YUM70, another selective HSPA5 inhibitor, attenuated tumor growth in a pancreatic cancer xenograft model (Samanta et al, 2021).Therefore, our preclinical in vivo studies strongly suggest HSPA5 inhibition as a promising treatment for NMSC.This hypothesis is supported by the high expression of HSPA5 in most NMSCs (https://www.proteinatlas.org/ENSG00000044574-HSPA5/pathology/skin+cancer), in particular in areas where strong downregulation of NRF3 is observed (this study).
In summary, we identified a tumor-suppressive function of NRF3 in the skin, which involves its interaction with the UPR regulator HSPA5.This offers promising therapeutic opportunities in stratified cancer patients.

Human skin and skin cancer samples
Surplus biopsies from normal human skin, AK, BCC, and SCC were obtained from the biobank of the Department of Dermatology, University Hospital Zurich (KEK Nr. 647/800).The diagnosis of AK and tumor samples was confirmed by a board-certified dermatopathologist (RD).Only samples from patients, who had signed an informed consent, were used in this study.The experiments were conducted in conformity to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report.The use of samples for research purpose was approved by the local ethics commission (KEK Nr. 2017-00684).

Genetically modified mice
Nrf3-ko mice (in C57BL/6 background) were previously described (Derjuga et al, 2004).NOD-SCID mice (NOD.CB17-Prkdcscid/ NCrCrl) were originally purchased from Jackson Laboratories (Bar Harbor, ME) and bred in the EPIC mouse facility at ETH Zurich.Mice were housed and fed according to federal guidelines, and all animal experiments had been approved by the local veterinary authorities of Zurich or Lausanne, Switzerland.Animals in tumor experiments were checked twice per week for the development of skin tumors.They were euthanized according to animal welfare regulations when a single tumor had a diameter of more than 1 cm.Mice were randomly assigned to the different treatment groups.

UVB irradiation of mice
Seven-to eight-week-old mice were irradiated with 100 mJ/cm 2 UVB exactly as previously described (Sch€ afer et al, 2010) and sacrificed 24 h post UV irradiation.Irradiation was monitored with a UV dosimeter.
DMBA/TPA treatment of mice 8-10 week-old female Nrf3-ko and wt mice received a single topical subcarcinogenic dose of 7,12-dimethylbenz(a)anthracene (DMBA; 100 lg in 300 ll acetone), followed by 25 weekly topical treatments with the tumor promoter 12-O-tetradecanoylphorbol-13-acetate (TPA; 15 lg in 200 ll acetone).Mice were sacrificed 30 or 46 weeks after DMBA treatment or when a tumor reached a size of 1 cm diameter or when more than 1 tumor with a size of more than 0.5 mm appeared, according to animal welfare regulations.

EMBO Molecular Medicine
Selina Gurri et al

Ear tumorigenicity assay
For xenograft tumor experiments, 2 × 10 5 cancer cells in 3 ll PBS were injected intradermally into the ear skin of 10-15 weeks-old NOD-SCID mice using a 33-gauge Hamilton micro-syringe.The control cells were injected into one ear and the cells under investigation into the other ear of the same mouse.Cells with inducible expression of a protein were pretreated for 24 h with 100 ng/ml Dox (Sigma-Aldrich; St. Louis, MO; D9891-5G) or vehicle (DMSO).Mice were then monitored for 3-4 weeks, and the size of the developing tumors was measured during the following 3-4 weeks.During the monitoring and measuring period, 20 mg/kg Dox was injected intraperitoneally (i.p.) every 3-4 days when cells with inducible overexpression were used.For the rescue experiment with the HSPA5 inhibitor (HA15), vehicle (DMSO) or HA15 (1.69 × 10 À3 mg/ear) was directly injected into the ear tumors every 2-3 days, for a total of 6 times.
Tumor length, width, and thickness were recorded every 2-3 days.Mitutoyo 7301 dial gauge (Mitutoyo, Urdorf, Switzerland) was used for thickness measurements; length and width were determined with a regular ruler.Before the tumors reached > 1 cm, the mice were sacrificed and the tumors harvested, bisected, and embedded in tissue freezing medium (Leica Microsystems, Wetzlar, Germany).
The sgRNAs were cloned into the lentiCRISPRv2 vector by restriction enzyme cloning as previously described (Sanjana et al, 2014).

Generation of NRF3 overexpression cells
The NRF3 cDNA was amplified and cloned into the pENTR/D-TOPO vector (Thermo Fisher Scientific, Waltham, MA; K240020) using the primers listed in Table 2. Subsequently, the LR Clonase II Enzyme mix (Thermo Fisher Scientific; 11791019) was used for cloning into the pInducer20 vector (kindly provided by Dr. S. Elledge, Harvard Medical School, Boston, MA) according to the manufacturer's protocol.

Cloning of an NRF3-BioID2 fusion protein
The Gibson Assemblyâ Cloning Kit (New England Biolabs [NEB]; Ipswich, MA; E5510S) was used to insert the NRF3 cDNA into the pENTR/D-TOPO vector containing a C-terminally HA-tagged BioID ligase using primers listed in Table 3. Subsequently, the cDNA encoding the NRF3-BioID fusion protein was inserted into the pInducer20 vector using the LR Clonase II Enzyme mix (Thermo Fisher Scientific; 11791019).

Lentivirus production
HEK 293T cells were grown to 50% confluency in 10 cm petri dishes in DMEM/10% FBS.Cells were transfected using Lipofectamine 2000 (Thermo Fisher Scientific; 11668019) according to the manufacturer's protocol using the psPax2 (Addgene; 12260) and pCMV-VSV-G (Addgene; 8454) packaging plasmids and the plasmid of interest.The next day, transfection media were removed, and fresh DMEM/10% FBS/1% P/S was added.After 48 h, the conditioned media containing the produced lentiviruses were collected, centrifuged at 1,200 rpm for 5 min to remove cell debris, and filtered through a 0.45 lm filter (Sarstedt, N€ urnbrecht.Germany; 83.1826.001).

Lentiviral transduction of cells
Cells were seeded into 6-well plates, incubated overnight at 37°C, 5% CO 2 , and transduced the next day.For lentiviral transduction, the virus-containing media were added to the cells in a 1:3 ratio with fresh DMEM/10% FBS/1% P/S.48 h after transduction and incubation at 37°C and 5% CO 2 , the virus-containing medium was removed and replaced by the selection medium, which contained either 5 mg/ml G418 (Thermo Fisher Scientific; 11811-031) or 1 lg/ml puromycin (P8833, Sigma-Aldrich) in DMEM/10% FBS.The medium was changed every second day, and the cells were assessed under the microscope.A well with non-transduced cells and cultured with the same selection medium served as control.Once all cells in the control wells had died and the transduced cells had   recovered, the surviving colonies were slowly expanded and maintained in selection medium to prevent overgrowth by cells, which had lost expression of the transgene.

Single cell clone expansion
For clonal expansion, cells were resuspended in 1 ml of 1% BSA in PBS containing propidium iodide (PI; Sigma-Aldrich; P4170; 1:1,000 diluted).A negative control without PI was also prepared.Cells were sorted by fluorescence activated cell sorting (FACS), and 1 cell per well was added into a 96-well plate with 200 ll selection medium.The 96-well plates were incubated at 37°C and 5% CO 2 , and the cells were slowly expanded.The purity of the different single cell clones was assessed by qRT-PCR and Western blot.

Cell treatment
Cells were seeded and treated on the following day.For inducible gene expression, the medium (DMEM/10% FBS/1% P/S) was supplemented with 100 ng/ml Dox and left on the cells for 24 h unless indicated differently.To induce ER-stress, cells were treated with 1 lg/ml tunicamycin (Sigma-Aldrich; T7765) for 24 h.To inhibit HSPA5, 25 lM HA15 (MedChemExpress, Princeton, NJ; HY-100437) or 5 lM YUM70 (MedChemExpress; HY-138364) were added to the cells for 24 h.As control, cells were treated with the same volume of vehicle (DMSO) for the same time period.

Preparation of protein lysates and Western blot (WB) analysis
Cells were washed with PBS, lysed in 240 mM Tris/HCl pH 6.8/ 280 mM SDS/40% (vol/vol) glycerol, and heated to 95°C for 10 min.After short centrifugation, the protein concentration was determined using the BCA Protein assay kit (Thermo Fisher Scientific; 23225) and a GloMax Microplate Reader (Promega, Madison, WI).The samples were diluted to 1 lg/ll in lysis buffer, and 10 mM dithiothreitol and bromophenol blue were added.Proteins (20 lg protein in total) were separated by SDS-PAGE and transferred to Protran nitrocellulose membranes (GE Healthcare; Chicago, IL; 10600001).Membranes were incubated for 1 h with 5% (wt/vol) skim milk powder (Rapilait; Migros, Switzerland; 150141000000) in Tris-buffered saline with Tween 20 (TBS-T) containing 25 mM Tris pH 8, 137 mM NaCl, 2.7 mM KCl, and 0.1% (vol/vol) Tween 20, to block unspecific binding sites.Cells were incubated with the primary antibody (see Table 4) diluted in 5% (wt/vol) BSA in TBS-T overnight at 4°C.The next day, the membranes were washed 3× for 10 min with TBS-T, incubated for 1 h with the secondary antibody (see Table 5) diluted in 5% (wt/vol) skim milk powder in TBS-T, and washed 3× for 10 min with TBS-T.Signals were developed using WesternBright Sirius HRP substrate (Advansta, San Jose, CA).

Co-Immunoprecipitation (co-IP)
Lysate from one 10 cm dish was used per precipitation.Cells were washed once with PBS, scraped off with 500 ll ice-cold PBS, and transferred to an Eppendorf tube.The plate was re-rinsed with 500 ll ice-cold PBS.Cells were centrifuged for 5 min at 500 g and 4°C.The supernatant was removed, and the cells were lysed with 500 ll IP buffer containing 0.05% NP-40, 10% glycerol, 150 mM NaCl, 5 mM HEPES, pH 7.4, 5 mM EDTA, and 0.25% deoxycholate, supplemented with PhosSTOP TM Phosphatase Inhibitor Cocktail (Sigma-Aldrich) and cOmplete TM EDTA-free Protease Inhibitor Cocktail (Sigma-Aldrich).Samples were incubated on ice for 10 min and centrifuged for 10 min at 16,000 g, 4°C.The supernatant was transferred to a new tube.50 ll were put aside as input fraction.25 ll

RNA isolation and qRT-PCR
Cells were washed with PBS, and RNA was extracted using the Mini Total RNA Kit (IBI Scientific, Dubuque, IA; IB47300) according to the manufacturer's protocol.1 lg RNA was reverse transcribed with the iScript cDNA synthesis kit (Bio-Rad, Hercules, CA; 1708890).The cDNA was diluted 1:10 in water.For the qPCR reaction, 5 ll diluted cDNA were mixed with 5.5 ll LightCycler SYBR green (Roche, Rotkreuz, Switzerland; 04887352001) and 0.5 ll primer mix (10 lM; Table 6).qRT-PCR signals were detected with a Light Cycler 480 II (Roche), and the data were analyzed using the double d cycle threshold (CT) method.
5-Bromo-2 0 -deoxyuridine (BrdU) assay Cells were seeded into 12-well plates containing one coverslip per well.The next day, BrdU (10 lM final concentration) was added to the cells, which were then incubated for 6 h at 37°C and 5% CO 2 .Afterward, they were washed once with 1× PBS and fixed with 4% PFA for 20 min.The DNA was denatured with 2 M HCl/0.1% Triton X-100 for 30 min.After addition of BrdU neutralization solution containing 0.1 M sodium borate pH 8.5 and incubation for 5 min, unspecific binding sites were blocked with 1% BSA in PBS for 1 h.Cells were incubated with the primary antibody (a-BrdU-FITC, Roche; 1:500 diluted) overnight at 4°C.Nuclei were then counterstained with Hoechst 33342 (Thermo Fisher Scientific; H3570).

Clonogenic assay
Totally, 10,000 SCC or HaCaT-Ras cells were seeded into each well of a 6-well plate and cultured at 37°C, 5% CO 2 for 7 days.Cells were then washed once with PBS, fixed and stained with 6% glutaraldehyde and 0.5% crystal violet in water for 30 min at room temperature (RT).After gently rinsing the wells with tap water, they were left to dry at RT. Plates were scanned with an Epson Perfection V700 Photo Scanner (Epson Deutschland GmbH, Meerbusch, Germany) and analyzed with ImageJ using the plugin "Colony Area" and the macro "Colony Measurer" (Fiji/ImageJ, Open Source).

Scratch assay
Cells were seeded into 6-well plates and cultured until they reached full confluency.To inhibit proliferation, 2 lg/ml mitomycin C (Sigma-Aldrich; M4287) was added for 2 h.Afterward, one scratch per well was made with a 200 ll pipette tip.The mitomycin Ccontaining medium was removed, and the wells were washed once with PBS before fresh medium was added.Pictures of the scratch were taken at different time points always at the same spot.They were analyzed manually or with the MRI wound healing tool (https://github.com/MontpellierRessourcesImagerie/imagej_macros_and_scripts/wiki/Wound-Healing-Tool).

Thiazolyl blue tetrazolium bromide (MTT) assay
Totally, 200,000 cells per well were seeded into a 12-well plate.The next day, cells were treated with DMSO (vehicle), HA15 (25 lM), or YUM70 (5 lM) for 24 h.Afterwards, 150 ll MTT (Abcam; ab146345) solution (5 mg/ml MTT in PBS) was added to each well and incubated for another 30 min.The supernatant was aspirated, and the cells were lysed with 225 ll MTT lysis buffer (49.5 ll HCl in 15 ml isopropanol) for 10 min at RT. 125 ll lysate was transferred to a 96-well plate, and the absorption was measured at 600 nm using the GloMaxâ Microplate reader (Promega).

Lactate dehydrogenase (LDH) cytotoxicity assay
Cells were seeded into 96-well plates.The LDH assay was performed 24 h after seeding using the Pierce TM LDH Cytotoxicity Assay Kit (Thermo Fisher Scientific) according to the manufacturer's protocol.Colorimetric measurements at 490 and 600 nm were obtained using a GloMaxâ Microplate reader (Promega).

Spheroid formation assay
Drops of 20 ll containing 2,000 cells were placed on the inner part of the lid of a 6 cm dish, which was then turned upside down.To prevent dehydration, 2 ml PBS was added to the bottom of the plate.Cells were incubated in the hanging drops with different cell culture media at 37°C and 5% CO 2 for 72 h.Pictures were taken, and the size of the spheroids was measured using ImageJ.

3D organotypic skin cultures
Scaffold-free organotypic 3D cultures were established as previously described (Berning et al, 2015).For this purpose, 1.86 × 10 6 human primary foreskin fibroblasts were seeded into 6-well inserts with a 0.4 lm transparent polyethylene terephthalate (PET) membrane in 1.5 ml of sterile-filtered Medium A (DMEM with high glucose/ L-glutamine and DMEM/F-12 GlutaMAX TM in a 1:1 ratio), supplemented with 0.2 mg/ml 2-phospho-L-ascorbic acid (Sigma-Aldrich; 49752), 1 ng/ml transforming growth factor b1 (Sigma-Aldrich;-T7039), 2.5 ng/ml epidermal growth factor (Sigma-Aldrich; E4127), 5 ng/ml fibroblast growth factor 2 (PeproTech, Cranbusy, NJ; 100-18B), and 5 lg/ml insulin (Sigma-Aldrich; I5500).Outside of the insert, another 12 ml of Medium A was added to Falconâ 6-well Deep Well tissue culture-treated polystyrene plates (Corning, New York, NY; 355467).Two additional rounds of fibroblast seeding as described above were performed after 2 and 4 days.Medium A inside and outside the 6-well insert was replaced every second day until day 31.On day 32, 10 6 SCC13 cells (NRF3-KO cells and their corresponding controls) were seeded on top of the dermal layer with 1.5 ml sterile-filtered Medium B (DMEM with high glucose/ L-glutamine and DMEM-F12 GlutaMAX TM in a 1:1 ratio, supplemented with 0.2 mg/ml 2-phospho-L-ascorbic acid, 0.0084 lg/ml cholera toxin [Sigma-Aldrich; C8052], and 0.4 lg/ml hydrocortisone [Sigma-Aldrich; H0888]).Additionally, 12 ml of Medium B was added to the well.After 3 days, the insert was air-lifted by removing the media on the insert and only replacing the outside medium with 12 ml of fresh Medium B. Cultures on the inserts were then incubated at the air-liquid interface for 4 weeks.The outside medium was replaced every second day with 12 ml of Medium B until collection (day 60).Mature cultures were harvested, bisected, and embedded in tissue freezing medium (Leica Microsystems).

Immunofluorescence and immunohistochemistry stainings
Sections were deparaffinized and rehydrated or fixed with ice-cold acetone (in case of frozen sections).If needed, antigen retrieval was performed prior to the blocking procedure by incubation in hot citrate buffer (1 h at 95°C).For immunohistochemistry, endogenous peroxidase activity was quenched with 3% H 2 O 2 for 10 min prior to blocking.Unspecific binding sites were blocked with 12% BSA for 1 h at 4°C, and sections were then incubated overnight at 4°C with the primary antibodies (see Table 7) diluted in the same buffer.

Hematoxylin and eosin (H&E) staining
Samples were stained with hematoxylin and eosin according to the protocol described in Table 9.Sections were mounted with Eukittâ (Merck, 03989) mounting medium.

Toluidine blue staining
Cells were stained with toluidine blue staining solution (Sigma-Aldrich; 198161) for 30 min at RT. Afterwards, they were washed twice for 5 s with ddH 2 O and counterstained with hematoxylin.

BioID labeling and sample preparation for mass spectrometry analysis
BioID interactome identification was adapted from Hesketh et al (2017).SCC13 and HaCaT cells with Dox-inducible expression of the NRF3-BioID-HA fusion protein were seeded into 15 cm dishes.The next day, plates were treated with 50 lM biotin (Sigma-Aldrich; B4501) and 1 lg/ml Dox or DMSO for 24 h.The following day, half of the samples were additionally treated with 10 lM MG-132 (Merck; 474790) for 6 h.Cells were scraped off with PBS, pelleted, and stored at À80°C until further usage.Cell pellets were thawed and lysed in a 4:1 volume:weight ratio in BioID lysis buffer (50 mM Tris pH 7.5, 150 mM NaCl, 0.4% SDS, 1% NP-40, .100 ll trypsin solution (1 lg trypsin in 100 ll ABC solution) was added to the beads and incubated under rotation overnight at 37°C. 100 ll of bead supernatant was collected, and the beads were vortexed in 100 ll Milli Q water.The 100 ll supernatants were pooled with the previously collected supernatants, and 50 ll formic acid was added.After evaporation on a SpeedVac, the dried peptides were dissolved in 20 ll 3% acetonitrile, 0.1% formic acid buffer and vortexed extensively.After sonication for 5 min, samples were placed on a thermomixer at 600 rpm for 10 min.

Mass spectrometry analysis of biotinylated proteins
Dissolved peptides were transferred to autosampler vials for liquid chromatography-mass spectrometry (LC-MS/MS) analysis.1 ll was injected on a NanoAcquity HPLC coupled to a Q-Exactive HF mass spectrometer (Thermo Fisher Scientific).Raw files were converted to Mascot Generic File formats (.mgf files) from which they were further analyzed using the Mascot search engine (Matrix Science, London, UK).For visualization of the protein and peptide identification results, Mascot protein identification results were imported into the Scaffold software (Proteome Software, Portland, OR).The data was uploaded to the Reprint online tool and analyzed using the CRAPome database.

RNA-sequencing
SCC13 NRF3-KO cells, which had been transduced with the pINDUCER20 or NRF3-pINDUCER20 vector, were incubated with 20 ng/ml Dox or DMSO (vehicle) for 10 h.Cells were washed with PBS, and RNA was extracted using the Mini Total RNA Kit (IBI Scientific; IB47300), according to the manufacturer's protocol.Using a 4200 TapeStation (Agilent Technologies, Santa Clara, CA), RNA quality was assessed, and libraries were prepared according to the Illumina TruSeq Stranded mRNA protocol.Libraries were pooled equimolarly and sequenced (single-end 100 bp) using an Illumina NovaSeq6000 instrument with a depth of approximately 20 Mio reads per sample.
For sequencing of mouse epidermal RNA, wt and Nrf3-ko mice were mock-treated or UVB-irradiated and sacrificed 24 h later.The epidermis was isolated exactly as described previously (Siegenthaler et al, 2018) and the RNA extracted with Trizol followed by purification with the RNeasy Mini Kit, including on-column DNase treatment (Qiagen, Hilden, Germany).The sequencing platform was Illumina HiSeq2500 (Illumina Inc., San Diego, CA).Preparation of the library was performed with Illumina's TruSeq Stranded mRNA kit with 700 ng total RNA as input.

RNA-Seq data quantification and statistical analysis
The raw reads were first cleaned by removing adapter sequences and poly-x sequences (> 9 nt used for detection) using fastp (Version 0.20.0;Chen et al, 2018).Sequence pseudo alignment of the resulting high-quality reads to the human reference genome (build GRCh38.p13), and quantification of gene-level expression (gene model definition from GENCODE release 37) was carried out using Kallisto (Version 0.46.1;Bray et al, 2016).To identify differentially expressed genes, we used the glm approach implemented in the software package DESeq2 (R version: 4.2.0,DESeq2 version: 1.36.0;Love et al, 2014).Gene expression differences with a P-value ≤ 0.01 and log 2 ratio ≥ 0.5 (N = 3) were considered significant.
For the mouse epidermis RNA-seq, the raw reads were first cleaned by removing adapter sequences, trimming low quality ends, and filtering reads with low quality (phred quality < 20) using Trimmomatic (Bolger et al, 2014).Sequence alignment of the resulting high-quality reads to the mouse reference genome (build GRCm38 with the gene annotations downloaded on 2014-02-25 from Ensembl), and quantification of gene level expression was carried out using RSEM (B.Li & Dewey, 2011).Differential expression was computed using the generalized linear model implemented in the Bioconductor package DESeq2 (Love et al, 2014).Genes showing altered expression with adjusted (Benjamini and Hochberg method) P-values < 0.05 (N = 3) were considered differentially expressed.

Ingenuity pathway analysis
The list of putative interaction partners found in the BioID screen with a Saint Probability of > 0.95 was used for Ingenuity pathway analysis (Qiagen).Data tables associated with the canonical pathways identified in the SCC13 or HaCaT cell line were exported to identify pathways in which the partners of NRF3 seem to be involved depending on their activation z-scores.

Statistical analysis
For all experiments, we used the largest possible sample size, and there was no exclusion of any animal or data point.Quantifications were performed blinded by the investigators.Statistical analyses were performed using the Prism 9 software (GraphPad Software, San Diego, CA).For comparison of two groups, non-parametric Mann-Whitney U test was performed; for comparison of more than two groups, Kruskal-Wallis test or twoway ANOVA and Bonferroni's multiple comparisons test were performed.Nonsignificant (ns): P > 0.05; *P < 0.05, **P < 0.01, ***P < 0.001 ****P < 0.0001.
Basal and squamous cell carcinomas of the skin are by far the most frequent types of cancer in humans, and their incidence is continuously increasing.There is a strong need for the development of more efficient, noninvasive treatments, which requires a thorough understanding of the underlying mechanisms and the identification of new targets for pharmacological intervention.

Results
We identified NRF3 as a potent tumor-suppressive protein in the skin.The invasively growing cancer cells of skin cancer patients loose this protein, which promotes their invasive migration and survival in the harsh tumor environment.Mechanistically, NRF3 was shown to reduce the amounts of the chaperone heat shock protein A5 (HSPA5) and its activity in the unfolded protein response.Pharmacological inhibition of HSPA5 or reduction in its levels rescued the high malignancy of NRF3-deficient cancer cells in 2D and organotypic 3D cell cultures and in preclinical mouse models.

Impact
These results suggest pharmacological HSPA5 inhibition as a promising therapeutic strategy in skin cancer patients, whose skin tumor cells have lost the expression of NRF3.
Expanded View for this article is available online.
next determined the tumorigenic potential of the SCC13 NRF3-KO cells in a xenograft mouse model.Control (EV C1 and EV C2) cells were injected into the left ear and the KO cells (KO C3 and KO C4) into the right ear of immunodeficient NOD/SCID mice.The tumors formed by NRF3-KO cells were larger compared to those formed by control cells as reflected by the larger tumor volume over time and the higher tumor weight at the day of sacrifice (Fig 4A and B).Histologically, the NRF3-deficient tumors showed high malignancy and invasive growth, resulting in penetration of the cancer cells through the ear cartilage (Fig 4B and C).Ki67 staining revealed increased tumor cell proliferation in NRF3-deficient tumors (Fig 4D).Similar results were obtained with NRF3-deficient HaCaT-Ras cells (Fig EV3G-J).Overexpression of NRF3 in SCC13 cells did not affect tumor growth in the xenograft model (Appendix Fig S2A-D).
lamin A/C.G Quantification of the area of open scratch at 9 h post scratch wounding in percentage of the original scratch area in SCC13 RescueNRF3 cells treated with 20 ng/ml Dox or vehicle (DMSO).N = 4-5.H Left: Area of spheroids formed by SCC13 EV and NRF3-KO cells in single hanging drops.Right: Representative pictures of the spheroids.N = 70-71 hanging drops.I Area of spheroids formed by SCC13 RescueNRF3 cells treated with 20 ng/ml Dox or vehicle (DMSO).N = 18 hanging drops.J Representative images of H&E-stained sections from 3D skin equivalents formed by primary human skin fibroblasts and either EV or NRF3-KO cells.Scale bar: 100 lm.D, dermal equivalent; E, epidermal equivalent.K Representative images of sections from SCC13 EV and NRF3-KO organotypic cultures immunostained for keratin 14 (K14; green) and collagen IV (red) combined with Hoechst staining (blue), and quantification of the K14-positive area below the basement membrane (BM).N = 3 independent cultures, n = 5 sections.Scale bar: 100 lm.Data information: Graphs show mean AE SD.P-values were determined using Mann-Whitney test (G, H, I, K) or Kruskal-Wallis (B-D).Source data are available online for this figure.

Figure 4 .
Figure 4. Loss of NRF3 enhances tumor formation and invasive growth of SCC13 cells in a xenograft mouse model.A Representative photographs of ~8-week-old tumors (indicated by arrows) formed in NOD/SCID mice upon intradermal injection of SCC13 EV and NRF3-KO cells and tumor volume at different time points of tumor development.N = 6 tumors per cell line.Graph shows pooled results from two control (EV C1 and EV C2) and two KO clones (EV C3 and EV C4).§ indicates time points when a mouse had to be sacrificed because the tumor had reached a size of almost 1 cm.B Representative H&E stainings of sections from tumors formed by SCC13 EV or NRF3-KO cells at day 59, respectively day 55.Scale bar: 1,000 lm.Graph shows tumor weight at endpoint.N = 6 per cell line.C Graph showing the percentage of tumors that had invaded through the basement membrane (N = 6) and a representative image of a mouse with an invasive tumor formed by the SCC13 NRF3-KO cells and histological staining of the invasive front (asterisk indicates cartilage; arrow indicates site of invasion).Scale bar: 50 lm.D Quantification of Ki67/E-cadherin-positive cells in the xenograft tumors (N = 6 tumors [3 per cell line, n = 5-15 sections]) and representative images of sections from tumors formed by SCC13 EV and NRF3-KO cells stained for Ki67 (green), E-cadherin (red), and counterstained with Hoechst (blue).White lines indicate the borders to the stromal area.Scale bar: 50 lm.Data information: Graph in (A) shows mean AE SEM.Graphs in (B) and (D) show mean AE SD.P-values were determined using 2-way ANOVA (A), Mann-Whitney U test (B, D) or Fisher's exact test (C).Black or gray dots indicate data points from different EV cell lines; dark or light blue dots indicate data points from different KO cell lines.Source data are available online for this figure.
Graphs show mean AE SD.N = 3. Mean expression in DMSO-treated cells was set to 1. C BioID proximity labelling strategy to identify potential NRF3 binding partners in SCC13 and HaCaT cells with Dox-inducible expression of HA-tagged NRF3-BioID fusion proteins.Created with BioRender.com.D Venn diagram showing common putative NRF3 interactors in SCC13 and HaCaT cells.E Table showing the 34 proteins that were identified as putative NRF3 interactors in SCC13 and HaCaT cells in the BioID screen and their respective Saint Probability (SP) scores.The small MAF proteins, which are known interactors of NRF3 in the nucleus, are highlighted in orange.F Ingenuity pathway analysis showing the pathways in which the putative interaction partners are involved (top) and String analysis showing that many of these proteins also interact with each other (bottom).G Co-immunoprecipitation of the HA-tagged NRF3-BioID fusion protein with Dvl2, HSPA5 and PDIA3 in transduced SCC13 cells cultured in the presence or absence of Dox (1 lg/ml) and in the presence of MG-132 (10 lM).GAPDH was used as loading control for the input and the unbound fraction.H Top panel: Proximity ligation assay using transduced SCC13 cells expressing the HA-tagged NRF3-BioID fusion protein and antibodies against the HA epitope and PDIA3.Scale bar: 100 lm.Middle panel: Proximity ligation assay using SCC13 wt cells and antibodies against NRF3 and PDIA3.Scale bar: 50 lm.Lower panel: Proximity ligation assay using SCC13 wt cells and antibodies against NRF3 and HSPA5.Scale bar: 100 lm.Source data are available online for this figure.

▸Figure 6 .
Figure 6.HSPA5 levels are enhanced in NRF3-KO cells and in invasively growing skin cancer cells.A Percentage of cleaved caspase 3 (cl Casp3) positive cells after treatment of SCC13 EV and NRF3-KO cells with 1 lg/ml tunicamycin for 24 h.N = 6 cultures.B Western blot analysis of total lysates from SCC13 EV and NRF3-KO cells treated with DMSO or 1 lg/ml tunicamycin for 24 h, using antibodies against NRF3, NRF1, HSPA5, cl PARP, CHOP, and tubulin (loading control) and quantification of the signal intensity for HSPA5, cl PARP, and CHOP.N = 6 (3 per cell line).C Top: Western blot of total lysates from SCC13 EV and NRF3-KO cells for HSPA5 and tubulin.Bottom: Quantification of the Western blot signal intensity (left) and qRT-PCR of RNA from SCC13 EV and NRF3-KO cells for HSPA5 relative to RPL27 (right).Mean expression in EV cells was set to 1. N = 6 (3 per cell line).D Representative images of sections from organotypic skin cultures with SCC13 EV and NRF3-KO cells stained for HSPA5 (red) combined with Hoechst staining (blue) and quantification of the HSPA5 staining intensity.N = 3 independent cultures, n = 3-5 sections.Scale bar: 50 lm.E Representative confocal images of sections from human BCC (upper two panels) and SCC (lower two panels) stained for NRF3 (green) and HSPA5 (red).Nuclei were counterstained with Hoechst (blue).Scale bar: 50 lm.Data information: Graphs show mean AE SD.P-values were determined using Kruskal-Wallis test (A, B) or Mann-Whitney U test (C, D).Black or gray dots indicate data points from different EV cell lines; dark or light blue dots indicate data points from different KO cell lines.Source data are available online for this figure.

◀
Figure 7. HSPA5 is important for the tumor-suppressive effect of NRF3.A, B Area of spheroids formed by SCC13 EV and NRF3-KO cells in the presence of 25 lM HA15 (A), 5 lM YUM70 (B) or vehicle (DMSO).N = 24-30 hanging drops.C Area of spheroids formed by SCC13 EV and NRF3-KO cells transfected with scrambled (scr) or HSPA5 (si1 and si4) siRNAs.N = 24-31 hanging drops.D Quantification of the area of open scratch at 9 h post scratch wounding in percentage of the original scratch area in confluent SCC13 EV and NRF3-KO cells transfected with scrambled or HSPA5 siRNAs.N = 6.E Schematic representation of the HA15 treatment experiment.NRF3-KO and EV HaCaT-Ras cells were injected into the ears of NOD/SCID mice at day 0. Starting from day 15, vehicle (DMSO) or HA15 (1.69 × 10 À3 mg/ear) were directly injected into the tumors every 2-3 days for a total of 6 times.Tumors were harvested at day 30.Created with BioRender.com.F, G Left: Representative pictures of ~4-week-old tumors (arrows) formed in NOD/SCID mice by HaCaT-Ras NRF3-KO (F) or EV (G) cells after intratumoral injection of vehicle (left) or HA15 (right), and tumor volume at different time points.N = 6 tumors per condition pooled from two independent experiments.Right: Representative H&E stainings of sections from tumors formed by HaCaT-Ras NRF3-KO (F) or EV (G) cells treated with vehicle or HA15 at day 30.Scale bar: 800 lm.H Left: Quantification of Ki67/E-cadherin-positive cells in the xenograft tumors.N = 3 tumors per condition, n = 5 sections.Right: Representative images of sections from tumors formed by HaCaT-Ras NRF3-KO or control cells treated with vehicle or HA15 stained for Ki67 (green), E-cadherin (red), and counterstained with Hoechst (blue).Scale bar: 50 lm.I Left: Quantification of cleaved caspase 3 (cl Casp3)/E-cadherin-positive cells in the xenograft tumors.N = 3 tumors per condition, n = 3-5 sections.Right: Representative images of sections from tumors formed by HaCaT-Ras NRF3-KO or control cells treated with vehicle or HA15 stained for cl Casp3 (green), E-cadherin (red), and counterstained with Hoechst (blue).Scale bar: 50 lm.

Name Sequence 5 0
to 3 0 sgRNA1 Forward: CAC CGG GTA AAG ATC TAG GTC TAC G Reverse: AAA CCG TAG ACC TAG ATC TTT ACC C sgRNA2 Forward: CAC CGA AAC TTG AGT CAT CTA GGT G Reverse: AAA CCA CCT AGA TGA CTC AAG TTT C sgRNA4 Forward: CAC CGC ACC GGT TGA CAA TCA TAT G Reverse: AAA CCA TAT GAT TGT CAA CCG GTG C The areas indicated with a rectangle are shown at higher magnification in the indent.Scale bar: 100 lm,

Table 1 .
List of sgRNAs cloned into the lentiCRISPRv2 vector.

Table 2 .
Primers used for TOPO cloning.

Table 3 .
Primers used for Gibson Assembly.

Table 4 .
List of primary antibodies used for WB.

Table 5 .
List of secondary antibodies and other reagents used for WB.Thermo Fisher Scientific; 88836) per sample were washed 4× with 500 ll IP buffer without PhosSTOP and protease inhibitors and once with 500 ll complete IP buffer and resuspended in 160 ll complete IP buffer.50 ll washed bead suspension was added to each lysate and incubated overnight at 4°C under rotation.50 ll of the supernatant were put aside as unbound fraction.Beads were washed 3× with 500 ll TBS/0.05%Tween-20 and once with 500 ll ddH 2 O, resuspended in 100 ll Western blot lysis buffer +1:10 dithiothreitol and boiled for 10 min at 95°C.

Table 7 .
List of primary antibodies used for immunostaining.

Table 8 .
List of secondary antibodies used for staining.