Disturbances to protein homeostasis (proteostasis) can lead to protein aggregation and inclusion formation, processes associated with a variety of neurodegenerative disorders. DNAJB proteins are molecular chaperones that have been identified as potent suppressors of disease-related protein aggregation. In this work, a destabilised isoform of firefly luciferase (R188Q/R261Q Fluc; termed FlucDM) was overexpressed in cells to assess the capacity of DNAJBs to inhibit inclusion formation. Co-expression of all DNAJB proteins tested significantly inhibited the intracellular aggregation of FlucDM. Moreover, we show that DNAJB proteins suppress aggregation by supporting the Hsp70 (HSPA)-dependent degradation of FlucDM via the proteasome. The serine-rich stretch in DNAJB6 and DNAJB8, essential for preventing fibrillar aggregation, is not involved in the suppression of FlucDM inclusion formation. Conversely, deletion of the C-terminal TTK-LKS motif in DNAJB6 and DNAJB8, a region not required to suppress polyglutamine aggregation, abolished the ability to inhibit inclusion formation by FlucDM. Thus, our data suggest that DNAJB6 and DNAJB8 possess two distinct regions for binding substrates, one that is responsible for binding β-hairpins that form during amyloid formation and another that interacts with exposed hydrophobic patches in aggregation-prone clients.
Many age-related neurodegenerative diseases, including Alzheimer's disease, Huntington's disease and amyotrophic lateral sclerosis (ALS), are associated with the expression of aggregation-prone proteins or polypeptide fragments that oligomerise and deposit into inclusions. For example, the deposition of the tau protein (also known as MAPT) and amyloid-β peptide into intracellular and extracellular deposits, respectively, are pathological hallmarks of Alzheimer's disease (Bucciantini et al., 2002; Hardy and Selkoe, 2002; Haass and Selkoe, 2007; Iqbal et al., 2009). Proteins containing expanded polyglutamine (polyQ) repeats aggregate to form amyloid precursors and mature fibrils, which have been linked to Huntington's disease and various ataxias (Zoghbi and Orr, 2000; Chiti and Dobson, 2006). Although many diseases are characterised by the formation of amyloid fibrils, not all proteins aggregate through an ordered mechanism. For example, superoxide dismutase 1 (SOD1) mutations, which are causative of some familial forms of ALS, result in the formation of highly disordered and hydrophobic amorphous precipitates (Banci et al., 2007; Prudencio et al., 2009). In addition, recent studies have demonstrated that almost all cases of sporadic ALS and frontotemporal dementia share a common neuropathology of predominantly amorphous, intracellular deposits that contain TAR DNA-binding protein 43 (TDP43) (Neumann et al., 2006; Adachi et al., 2009; Scotter et al., 2015). Furthermore, some proteins are known to adopt intermediate conformations before forming amyloid fibres (Dobson, 2003; Stathopulos et al., 2003). Finally, amorphous aggregation is associated with protein misfolding that occurs during conditions of cellular stress (Chiti and Dobson, 2006; Ecroyd and Carver, 2008). Thus, protein aggregation can lead to amyloid fibril and/or amorphous forms, each with their own unique characteristics (Kampinga and Bergink, 2016). These aggregation-prone forms likely require different protein quality control mechanisms to maintain them in a soluble state.
Cellular protein homeostasis (proteostasis) is maintained by an interconnected protein quality control (PQC) network that helps to preserve a stable and functional proteome. The overload or failure of these PQC systems to maintain the solubility and function of aggregation-prone proteins has been hypothesised to result in the formation of inclusions and deposits associated with protein conformational diseases (Kampinga and Bergink, 2016). Molecular chaperones, in conjunction with various co-factors, including co-chaperones, are a key component of the PQC network. Molecular chaperones have a central role in ensuring the correct folding of nascent polypeptides, the maintenance of partially folded protein intermediates in a folding-competent state, the re-folding of damaged proteins and, where required, the shuttling of misfolded aggregation-prone proteins for degradation. The heat shock proteins (Hsps) are a family of evolutionarily conserved molecular chaperones. They were discovered as proteins whose expression and activity are dramatically upregulated in response to a variety of (mostly) proteotoxic forms of stress (Feder and Hofmann, 1999). Most of these stress-inducible Hsps bind to exposed hydrophobic regions of proteins and either assist in their refolding, or traffic irreversibly damaged proteins for proteolytic degradation via the proteasome or by autophagy (Powers et al., 2009; Morimoto, 2011).
Critical components of the chaperone network include the Hsp70 machinery (here referring to the HSPA family of Hsps). The activity of these Hsp70 chaperone machines is driven by interactions with DNAJ proteins, whereby DNAJs deliver misfolded substrates to the Hsp70 machinery and, in turn, stimulate Hsp70 ATPase activity (Bukau et al., 2006). The human genome encodes 53 DNAJ proteins that can be divided into three separate subfamilies, the DNAJA, DNAJB and DNAJC proteins, based upon intrinsic structural features (Cheetham and Caplan, 1998; Kampinga and Craig, 2010). Of these, the DNAJB proteins are the most extensively studied due to them being previously identified as potent suppressors of aggregation associated with many disease-related proteins, including polyQ, amyloid-β and SOD1 proteins. For example, previous work has shown that DNAJB2a and the closely related proteins DNAJB6a, DNAJB6b and DNAJB8 potently suppress the aggregation of a polyQ-expanded protein in a cell culture model of disease (Howarth et al., 2007; Hageman et al., 2010; Gillis et al., 2013). Increasing the expression of DNAJB2a or DNAJB6b in a mouse model of Huntington's disease delays polyQ aggregation, alleviates symptoms and prolongs lifespan (Labbadia et al., 2012; Kakkar et al., 2016b). Moreover, DNAJB6b, both in vitro and in cells, prevents the nucleation of the amyloid-β peptides into mature fibrils (Månsson et al., 2014). More recently, mutant SOD1 aggregation was reported to be significantly suppressed by DNAJB1, DNAJB2b and DNAJB7, whereas other DNAJBs were found to have little or no effect (Serlidaki et al., 2020).
Structurally, DNAJB proteins contain a highly conserved N-terminal J domain, an internal glycine/phenylalanine (G/F)-rich linker region and a C-terminal domain (Cheetham and Caplan, 1998). The J domain contains the conserved histidine-proline-aspartate (HPD) motif for interaction with the Hsp70 machinery, whilst the C-terminal region is thought to bind substrates (Kampinga and Craig, 2010). Specialised members of the DNAJB family (DNAJB6 and DNAJB8) contain a serine/threonine (S/T)-rich motif between the G/F-rich region and C-terminal domain. The hydroxyl groups within the side chains of this S/T-rich region participate in intramolecular hydrogen bonding with β-hairpin structures in amyloid-β and polyQ peptides to prevent their primary nucleation into mature (disease-causing) toxic amyloid fibres (Hageman et al., 2010; Månsson et al., 2014; Kakkar et al., 2016b). The role of the G/F-rich region is currently not well understood; however, mutations in this region have been linked to reduced substrate binding capacity (Perales-Calvo et al., 2010) and have been implicated in inheritable forms of limb girdle muscular dystrophy (Harms et al., 2012; Sarparanta et al., 2012). It has therefore been hypothesised that the G/F-rich region may also be directly involved in substrate binding, as well as participating as a flexible linker region for inter-domain stabilisation.
To date, the majority of cell-based studies conducted to reveal the interaction of DNAJBs with client proteins have investigated proteins whose aggregation is disease related (Hageman et al., 2010, 2011; Månsson et al., 2014; Kakkar et al., 2016a; Serlidaki et al., 2020). In this work, we have chosen to exploit a previously described mutant isoform of firefly luciferase (R188Q/R261Q Fluc; herein referred to as FlucDM; Gupta et al., 2011) to assess the capacity of DNAJB molecular chaperones to inhibit the aggregation of destabilised client proteins in cells. This FlucDM isoform acts as a proteostasis sensor by reporting on the capacity of cells to maintain aggregation-prone proteins in a soluble state. Moreover, because Fluc forms amorphous-type aggregates (Schröder et al., 1993; Buchberger et al., 1996; Rampelt et al., 2012), this enabled us to assess whether the mechanism used by DNAJBs to suppress the fibrillar aggregation of proteins (Månsson et al., 2014; Kakkar et al., 2016b) is also used to inhibit the amorphous aggregation of destabilised proteins in cells.
We report that co-expression in cells of DNAJBs with FlucDM leads to a significant decrease in inclusion formation by FlucDM. Interaction with Hsp70 is required for this activity, since mutations that prevent this interaction abolish the capacity of these DNAJBs to suppress the aggregation of FlucDM into inclusions in cells. Furthermore, our data suggest that DNAJBs support Hsp70-dependent FlucDM degradation via the proteasome. The S/T-rich domain in DNAJB6 and DNAJB8, which is essential for inhibiting nucleation of polyQ amyloid fibrils (Hageman et al., 2010; Kakkar et al., 2016b), is not required for suppression of FlucDM aggregation; however, the short TTK-LKS motif in the C-terminal of DNAJB6 and DNAJB8, a region that plays no role in the suppression of polyQ aggregation (Hageman et al., 2010), is essential for anti-aggregation activity against FlucDM. Taken together, our data suggest that DNAJB6 and DNAJB8 contain at least two regions for binding substrates, an S/T-rich domain for binding β-hairpin structures that go on to form amyloid fibrils and a TTK-LKS region that interacts with exposed areas of hydrophobicity in destabilised substrates. Thus, the intrinsic properties of an aggregation-prone client may dictate the mechanism by which these specialised DNAJBs interact with client proteins.
DNAJB family members are potent suppressors of FlucDM aggregation
We exploited a double mutant isoform of firefly luciferase (FlucDM) that readily forms amorphous aggregates in cells without the need to apply an unfolding stress (Gupta et al., 2011) to determine the capacity of individual DNAJB chaperones to inhibit the aggregation of destabilised proteins. We first confirmed the intracellular aggregation propensity of FlucDM into inclusions by transfecting cells so they expressed EGFP-tagged FlucDM (FlucDM–EGFP) or the less aggregation-prone wild-type Fluc (Fig. 1A). Some cells expressing FlucDM–EGFP contained green fluorescent puncta throughout the cytoplasm (∼15% of cells), corresponding to the aggregation and inclusion formation of this protein. Although inclusions were occasionally observed in cells expressing FlucWT–EGFP (less than 5% of cells), most of these cells exhibited diffuse green fluorescence throughout the cytoplasm and the nucleus.
The significantly enhanced aggregation propensity of FlucDM–EGFP compared to FlucWT–EGFP was confirmed by examining the distribution of detergent-insoluble protein in cells expressing these proteins (Fig. 1B). There was approximately triple the amount of insoluble protein detected in cells expressing FlucDM compared to that in cells expressing FlucWT. Cells were also analysed using the flow cytometric analysis of inclusions and trafficking (FloIT) assay, a technique that readily enumerates the number of inclusions formed in cells (Whiten et al., 2016). The number of inclusions measured by FloIT was significantly higher (∼3-fold) in cells expressing EGFP-tagged FlucDM compared to the number in cells expressing EGFP-tagged FlucWT (Fig. 1C), in accordance with the results from the fluorescence microscopy and detergent insolubility assays. Taken together, these data show that FlucDM readily aggregates in cells and that FloIT can be used as a rapid and non-subjective method to assess this aggregation.
Previous studies have identified that the fibrillar aggregation of polyQ-expanded proteins can be significantly reduced by DNAJB2a, DNAJB6b and DNAJB8, whereas other DNAJBs are much less effective (Hageman et al., 2010). To test the ability of DNAJBs to engage destabilised proteins at risk of forming amorphous aggregates in cells, V5-tagged DNAJB proteins were transiently co-expressed with EGFP-tagged FlucDM in Flp-In T-REx HEK293 cells. Using the FloIT assay to assess the aggregation of FlucDM, we found that inclusion formation was significantly reduced by all DNAJBs tested, with DNAJB1, DNAJB5, DNAJB6b and DNAJB8 being the most potent suppressors of inclusion formation (Fig. 2).
Our finding that all of the DNAJBs tested significantly suppressed FlucDM aggregation contrasts with the previously reported ability of only a few specific DNAJB isoforms (DNAJB2a, DNAJB6 and DNAJB8) to strongly inhibit polyQ aggregation, with DNAJB1 having an intermediate effect (Hageman et al., 2010). Furthermore, unlike what we observed for suppression of inclusion formation by FlucDM, DNAJB4, DNAJB5 and DNAJB9 have been shown to be significantly less active against polyQ aggregation, with DNAJB2b having no effect (Hageman et al., 2010). This suggests that the mechanism by which DNAJBs act to suppress the amorphous aggregation of FlucDM is not the same as that used to suppress the fibrillar aggregation of proteins. Because DNAJB6b (hereafter referred to as DNAJB6) and DNAJB8 (both strong polyQ aggregation inhibitors), as well as DNAJB1 (a weak polyQ aggregation inhibitor), were among the DNAJBs most effective at suppressing FlucDM aggregation, we focussed on these isoforms in order to further interrogate this mechanism of action.
DNAJBs promote the degradation of FlucDM, primarily via the proteasome
We sought to determine whether the inhibition of FlucDM aggregation into inclusions by DNAJB proteins requires the degradative activity of the proteasome or autophagy. To do so, HEK293 cells were co-transfected to express EGFP-tagged FlucDM and DNAJB1 or DNAJB6 (or mRFP as a control) and, 24 h post-transfection, were treated with proteasome (MG132) or autophagy (3-methyladenine plus bafilomycin A1) inhibitors and analysed at 48 h. Treatment of cells with inhibitors of autophagy had no significant effect on the level of inclusion formation and little effect on the insolubility of FlucDM in cells co-expressing a DNAJB (Fig. 3A,B). Inhibition of autophagy was confirmed in cells treated with 3-methyladenine plus bafilomycin A1 by the increased levels of SQSTM1 (also known as p62; a commonly used marker of autophagy). An increase in SQSTM1 expression was also observed in cells treated with MG132, an effect that has been reported previously whereby inhibition of the proteasome leads to upregulation of SQSTM1 transcription (Myeku and Figueiredo-Pereira, 2011), suggesting a crosstalk between these two pathways (Liu et al., 2016). As there was no substantial increase in the level of insoluble FlucDM in cells treated with the autophagy inhibitors, these data suggest that FlucDM is primarily degraded by the proteasome.
Upon treatment with MG132, the number of FlucDM inclusions, as assessed by FloIT, significantly increased in cells co-expressing the mRFP non-chaperone control (Fig. 3A), and this corresponded to an increase in the proportion of FlucDM found in the NP-40-insoluble fraction (Fig. 3B). Inhibition of the proteasome following treatment with MG132 was evidenced by large smears of polyubiquitylated protein in these samples. Proteasome inhibition also led to a significant increase in inclusion formation in cells overexpressing DNAJBs compared to that in DMSO-treated cells, such that the capacity of the co-expressed DNAJBs to reduce the amount of insoluble FlucDM was significantly reduced when cells were treated with MG132. Additionally, the amount of soluble FlucDM also decreased in cells expressing DNAJBs that were treated with MG132, as compared to the amount in DMSO-treated controls. This was despite aggregation still being significantly reduced in MG132-treated cells that overexpressed a DNAJB compared to that in MG132-treated cells expressing mRFP. These data suggest that DNAJBs can maintain destabilised FlucDM in a non-aggregated soluble form (proteasome independent) such that, with time, the proteasome can facilitate its degradation. Furthermore, FlucDM aggregation is dependent upon proteasomal degradation, and the inhibition of FlucDM aggregation into inclusions by DNAJBs requires the activity of the proteasome.
The J domain is crucial for DNAJBs to protect against FlucDM aggregation
We next examined whether DNAJBs require an interaction with Hsp70 in order to suppress the aggregation of EGFP-tagged FlucDM into inclusions. To do so, we employed mutant forms of the DNAJBs in which a histidine residue was replaced with a glutamine (H/Q) within the highly conserved HPD motif of the J domain (Hageman et al., 2010) (Fig. 4A). The HPD motif plays a critical role in the regulation of Hsp70 activity; the H/Q mutation in this motif blocks the ability of the DNAJB to interact with Hsp70 (Cheetham and Caplan, 1998), thereby abrogating its ability to stimulate Hsp70 ATPase activity (Tsai and Douglas, 1996) and recruit Hsp70 to clients. The H/Q mutation abolished the capacity of each of the three DNAJB proteins to inhibit the aggregation of FlucDM, as evidenced by FloIT and assessment of the aggregation of FlucDM through cell fractionation using NP-40 (Fig. 4B,C). Thus, co-expression of the wild-type DNAJBs reduced the amount of insoluble protein, whereas cells expressing the H/Q mutant isoforms contained an equivalent or increased amount of insoluble FlucDM compared to that in the mRFP control. The amount of soluble FlucDM in cells expressing a wild-type DNAJB decreased compared to the amount in cells expressing the mRFP control. This effect is likely a result of there being less total FlucDM in cells expressing wild-type DNAJBs, due to them promoting FlucDM degradation. The expression level of the DNAJB H/Q variants was slightly higher than that of the corresponding wild-type protein, and this could be due to the mutant proteins becoming trapped with their substrates within inclusions, such that their own normal turnover is delayed. Strikingly, the relative loss in activity of the H/Q variants was highest for DNAJB1 (i.e. largest increase in insoluble protein compared to the amount in cells expressing the wild-type variant), and the ratio of insoluble to soluble FlucDM in cells expressing DNAJB1 H/Q was different to that in cells expressing DNAJB6 H/Q or DNAJB8 H/Q.
We next investigated whether increased activation of heat shock factor 1 (HSF1) may account for the observed effect whereby DNAJB overexpression led to decreased FlucDM inclusion formation, for example by increasing levels of Hsp70. Thus, the expression of Hsp70 was also assessed following overexpression of these DNAJB isoforms, because levels of some Hsp70s increase when HSF1 is activated. As expected, there was an increase in the expression of Hsp70 in heat-shocked cells as a result of HSF1 activation (∼2 fold); however, the expression of Hsp70 was not increased in cells overexpressing DNAJBs compared to that in cells expressing mRFP as a control (Fig. 4C). Thus, these data indicate that increased activity of HSF1 does not account for the decrease in FlucDM inclusion formation in cells overexpressing DNAJBs. Moreover, given that the expression of Hsp70 was not affected by DNAJB overexpression, these data imply that, although DNAJBs prevent the aggregation of FlucDM by interacting with Hsp70, their mode of action and relative dependence on Hsp70 may be dissimilar.
DNAJB proteins facilitate interaction with Hsp70 and FlucDM for proteasomal degradation
In order to examine whether DNAJB proteins mediate FlucDM degradation by the proteasome via interaction with Hsp70, we co-expressed EGFP-tagged FlucDM and the DNAJB H/Q variants in cells and then treated them with MG132. We surmised that if Hsp70 is the driver of proteasomal degradation of FlucDM, the H/Q variants, which are unable to interact with Hsp70, should not further increase the levels of inclusions formed in MG132-treated cells. Inhibition of the proteasome in these experiments was again confirmed by an increase in polyubiquitylated proteins, observed as large, high molecular weight smears by immunoblotting with an anti-ubiquitin antibody. As before, there was a significant increase in the number of inclusions in cells expressing either the DNAJB1 H/Q or DNAJB8 H/Q variants compared to the number in mRFP-expressing control cells (Fig. 5A); however, inclusion formation did not further increase in cells expressing an H/Q variant upon treatment with MG132. Again, the result was different for DNAJB1 H/Q compared to DNAJB6 H/Q and DNAJB8 H/Q, whereby treatment of cells expressing DNAJB1 H/Q with MG132 led to a decline in the number of inclusions compared to that in the DMSO-treated control. Since MG132 is a substrate analogue (Lee and Goldberg, 1998), this effect could be attributed to the drug interfering with FlucDM–DNAJB1 H/Q complex formation. The C-terminus of DNAJB1, which is thought to be responsible for substrate binding, structurally differs from that of DNAJB6 and DNAJB8 (Kampinga and Craig, 2010) and this difference may explain why the effect observed was specific for DNAJB1 H/Q. There was no difference between the amount of insoluble protein detected in MG132-treated cells expressing the H/Q variants and DMSO-treated control cells expressing the H/Q variants (Fig. 5B). We did note some inter-assay variability for cells expressing mRFP treated with MG132 compared to previous experiments (Fig. 3A); we attribute this to differences in the amount of time cells were treated with MG132 (i.e. cells were treated with MG132 for 24 h in the experiments presented in Fig. 3A and for 18 h in the experiments presented in Fig. 5A). Taken together, these data provide further evidence that DNAJBs antagonise FlucDM aggregation by keeping it competent for proteasomal degradation, which requires interaction with Hsp70 to be effective.
Disease-related mutations in the G/F-rich region of DNAJB6 do not impact the capacity to prevent the aggregation of FlucDM into inclusions
To probe for other regions within DNAJB proteins that are required for suppression of FlucDM aggregation, we first assessed the impact of two disease-related missense mutations within the G/F-rich region of DNAJB6 (F93L and P96R; Fig. 6A). The F93L and P96R mutations have been associated with limb girdle muscular dystrophy, and it has been suggested that these mutations lead to disruption of the J-to-G/F interdomain interaction and minor loss of function in the capacity to suppress polyQ aggregation (Sarparanta et al., 2012; Thiruvalluvan et al., 2020). However, we found that both the F93L and P96R mutational variants of DNAJB6 fully retained the ability to inhibit the aggregation of destabilised FlucDM–EGFP into inclusions (Fig. 6B,C).
The C-terminus, and not the serine-rich region, of DNAJB proteins is required for suppression of FlucDM inclusion formation in cells
Previous work has suggested that the hydroxyl groups of serine/threonine (S/T) side chains in the C-terminal domain of DNAJB6 participate in intramolecular hydrogen bonding with polyQ peptides and that this likely mediates inhibition of amyloid formation. For example, increasing the number of S/T residues replaced with alanine (A) residues (from 6 to 13 to 18 substitutions; variants referred to as M1, M2 and M3, respectively; Fig. 7A) leads to a progressive loss in the ability of DNAJB6 to inhibit polyQ or amyloid-β aggregation, with the M3 variant being functionally inactive in these assays (Kakkar et al., 2016b; Månsson et al., 2018). Interestingly, we found that the DNAJB6 M1, M2 and M3 variants fully retained their ability to suppress intracellular inclusion formation of EGFP-tagged FlucDM (Fig. 7B,C), indicating that the hydroxyl groups of the S/T-rich domain of DNAJB6 are not required for interaction between DNAJB6 and FlucDM. Deletion of almost the entire S/T-rich region of the C-terminus of DNAJB6 (giving rise to a variant referred to as M4) did result in abrogation of DNAJB6-mediated suppression of FlucDM aggregation; however, this is likely due to the structural destabilisation of this DNAJB6 mutant, which results in it being readily degraded (Kakkar et al., 2016b), as was evidenced by its very low levels in the lysate from transfected cells (Fig. 7C). Taken together, these data imply that the residues in DNAJB6 responsible for the inhibition of the amorphous aggregation of FlucDM into inclusions differ from those used to suppress amyloid fibril-type aggregation of proteins.
To further define the functional regions within the C-terminal domain responsible for the anti-aggregation activity of DNAJBs, the DNAJB6 ΔTTK-LKS deletion construct was co-expressed with EGFP-tagged FlucDM in cells (Fig. 7A). Deletion of the short C-terminal TTK-LKS motif, which in DNAJB8 is dispensable for inhibiting polyQ aggregation (Hageman et al., 2010), did abrogate the capacity of DNAJB6 to inhibit FlucDM inclusion formation (Fig. 7D,E). Consistent with the data obtained for DNAJB6, deletion of the TTK-LKS motif in DNAJB8 also abrogated this activity (Fig. 7F,G), further indicating different requirements for dealing with amorphous FlucDM aggregation compared to polyQ aggregation. In line with the results obtained following expression of the M3 isoform of DNAJB6, deletion of the S/T-rich region (ΔSSF-SST) of DNAJB8 had no effect on the ability of DNAJB8 to suppress the aggregation of FlucDM.
Immunostaining and confocal microscopy were undertaken on HEK293 cells co-expressing EGFP-tagged FlucDM with either V5-tagged wild-type DNAJB6 or DNAJB8, or various mutational variants (or mRFP as a control). Cells expressing FlucDM–EGFP together with the mRFP (non-chaperone) control contained many punctate FlucDM inclusions located throughout the cytoplasm, whilst mRFP remained diffuse and was expressed in both the cytoplasm and nucleus (Fig. 8). Overexpression of DNAJB6 or DNAJB8 resulted in fewer cells with FlucDM inclusions. In those cells with FlucDM inclusions, the wild-type DNAJB proteins were colocalised with the inclusions. A greater proportion of cells expressing the DNAJB6 H/Q variant contained FlucDM inclusions, and the H/Q variant also colocalised with the inclusions. As expected, fewer cells that expressed the DNAJB6 M3 variant contained FlucDM inclusions, and colocalisation of FlucDM inclusions with DNAJB6 M3 was also observed. Conversely, deletion of the TTK-LKS region of DNAJB8 resulted in increased FlucDM inclusion formation compared to that in cells expressing wild-type DNAJB8, and DNAJB8 ΔTTK-LKS was not found to colocalise with FlucDM in inclusions. Taken together these data suggest that DNAJB6 and DNAJB8 may have at least two different regions involved in substrate handling, one that is responsible for proteins that form β-hairpins during amyloid formation (Kakkar et al., 2016b) and another that is required for the handling of destabilised aggregation-prone proteins, such as those represented here by FlucDM.
In this work, we demonstrate that the DNAJB molecular chaperones are potent suppressors of the aggregation of FlucDM into inclusions in cells. This contrasts with what has been observed previously, whereby only specific DNAJB isoforms suppress the aggregation of polyQ-expanded proteins (Hageman et al., 2010). We show that DNAJB1, DNAJB6 and DNAJB8 inhibit FlucDM aggregation in a manner that depends on their ability to interact with Hsp70 and is associated with the cellular capacity to degrade FlucDM via the proteasome, thereby alleviating protein aggregation. For DNAJB6 and DNAJB8, the suppression of FlucDM aggregation not only appears mechanistically different to that mediated by DNAJB1, but is also distinct from what has been reported previously for the handling of amyloid-fibril-forming proteins, such as polyQ-expanded proteins and the amyloid-β peptide (Hageman et al., 2010; Kakkar et al., 2016b; Månsson et al., 2018). Although the S/T-rich region and, to some extent, the G/F-rich region (Sarparanta et al., 2012; Thiruvalluvan et al., 2020) in DNAJB6 and DNAJB8 are essential for amyloid suppression, these regions are not required to inhibit FlucDM inclusion formation. Finally, we identified a short, 23-amino-acid (TTK-LKS) sequence in the C-terminus of DNAJB6 and DNAJB8 that is required to inhibit FlucDM inclusion formation; this region in these proteins is dispensable for suppression of polyQ aggregation (Hageman et al., 2010). Thus, whilst DNAJB6 and DNAJB8 are both potent inhibitors of polyQ and FlucDM aggregation, the mechanism by which they interact with these aggregation-prone proteins is different. Our data suggest that DNAJB6- and DNAJB8-like proteins possess distinct regions for interacting with clients and that this is likely dictated by the structure or composition of the aggregation-prone protein.
We exploited the previously described flow cytometry technique FloIT (Whiten et al., 2016) to screen the capacity of various DNAJB isoforms to suppress FlucDM inclusion formation. Previous to this, Hsp overexpression screens to identify inhibitors of protein aggregation have typically relied on using traditional bulk-based biochemical analyses, such as the filter trap assay or fluorescence microscopy (Hageman et al., 2010; Kakkar et al., 2016a; Serlidaki et al., 2020). However, when screening many different samples, these types of assays can be time consuming, laborious and, when it comes to counting inclusions in individual cells, subjective. Although the basic principle behind methods such as the filter trap assay and FloIT are the same (i.e. analysis of insoluble protein in a cell lysate), FloIT offers several advantages. First, FloIT can be used in a medium-to-high-throughput capacity for rapid non-subjective quantification of inclusions across samples that may differ in transfection efficiency and cell number. Second, FloIT can identify (and enumerate) inclusions that differ in size, granularity and even protein composition, making FloIT broadly applicable to most (if not all) model systems of protein aggregation in cells (Whiten et al., 2016). Here, we describe the first use of FloIT as a quantitative method to screen for the ability of molecular chaperones (and mutational variants) to prevent protein aggregation in cells, highlighting the power and potential applications of the technique for the study of proteostasis.
The S/T-rich stretch in DNAJB6 (amino acids 155–195) and DNAJB8 (amino acids 149–186) is highly conserved between these proteins. It has been proposed that interaction with hydroxyl groups in side chains of these S/T residues inhibits primary nucleation by outcompeting for hydrogen bonding essential for β-hairpin and mature amyloid fibril formation, thereby suppressing aggregation (Kakkar et al., 2016b). DNAJB2 also contains a partial serine-rich stretch and, although it is not confirmed to be involved in polyQ handling, is also more effective than DNAJB1 and DNAJB5 (which lack this region) at suppressing polyQ aggregation (Hageman et al., 2010). DNAJB5 has close homology to DNAJB1 (Chen et al., 1999) and has been shown to interact with Hsp70 (Hageman et al., 2011). Moreover, since DNAJB5 was also identified as a potent suppressor of FlucDM inclusion formation, it is likely that DNAJB1 and DNAJB5 inhibit FlucDM aggregation into inclusions via a similar mechanism. Mutation or deletion of the S/T-rich region did not result in loss of the ability of DNAJB6 or DNAJB8 to suppress the aggregation of FlucDM into inclusions, indicating that a different region of the protein is involved in this process. Interestingly, DNAJB1 and the other DNAJBs we tested were all capable of suppressing FlucDM aggregation. Whilst our data does not provide insight into the domains required by these other DNAJBs to prevent FlucDM aggregation, our work has identified a short TTK-LKS (TTKRIVENGQERVEVEEDGQLKS) fragment conserved between DNAJB6 (amino acids 204–226) and DNAJB8 (amino acids 195–217) that is crucial for the handling of FlucDM. Given that this TTK-LKS domain in DNAJB8 is dispensable for its capacity to inhibit polyQ aggregation (Hageman et al., 2010), our findings are the first to demonstrate that DNAJB6 and DNAJB8 have two distinct regions for handling client proteins. Little is currently known regarding the functional role of this conserved TTK-LKS motif in DNAJB6 and DNAJB8; however, based on our current data, we hypothesise that it is either directly or indirectly involved in binding hydrophobic patches in destabilised aggregation-prone proteins.
Recent structural homology modelling of the DNAJB6 dimer/oligomer revealed four β-strands within the C-terminal domain of DNAJB6 (Söderberg et al., 2018). Dimerisation of each DNAJB6 monomer likely occurs via same-to-same-residue crosslinks at lysine residues K189 and K232 within the first and fourth β-strands, respectively. When crosslinked to form a dimer, the symmetrically positioned β-strands within DNAJB6 monomers form a peptide-binding pocket that is surface exposed and lined with the S/T residues responsible for binding fibrillar proteins. Based on these structural data, the TTK-LKS region (which lies downstream of the S/T-rich region) is contained within the fourth β-strand of DNAJB6 (and of DNAJB8), which is surface-exposed in the monomeric and dimeric form of DNAJB6. Thus, this region has the potential to be a second substrate-binding region in DNAJB6 and DNAJB8, responsible for binding hydrophobic aggregation-prone client proteins.
One possible reason why DNAJB6- and DNAJB8-like proteins possess distinct mechanisms to interact with aggregation-prone client proteins is due to intrinsic structural differences in misfolded states of proteins that lead to the formation of amorphous aggregates as opposed to amyloid fibrils. PolyQ-expanded proteins form large, tightly aggregated structures that are extremely insoluble and typical of amyloidogenic deposits (Hageman et al., 2010; Kubota et al., 2011). The R188Q and R261Q mutations in the N-terminus of the Fluc variant used in this study conformationally destabilises the protein (Gupta et al., 2011), thereby inducing protein misfolding and increased regions of exposed hydrophobicity. This causes the protein to form aggregates that are SDS-soluble and localise into diffuse cytosolic inclusions (Gupta et al., 2011), distinct from the amyloid-like aggregates formed by polyQ-expanded proteins. Indeed, when both polyQ-expanded huntingtin and FlucDM are expressed together in human cell lines, the two proteins deposit into distinct aggregated structures (Gupta et al., 2011), reaffirming that they aggregate via different mechanisms. Importantly, our data highlight that it may be possible to design therapeutics that boost the ability of DNAJB6 and DNAJB8 to prevent amyloid fibril formation associated with disease, whilst not impacting the capacity to interact with highly destabilised aggregation-prone proteins destined for degradation by the proteasome.
A major finding of this work is that all of the DNAJBs tested significantly inhibited FlucDM inclusion formation, a result that contrasts with previous observations regarding the suppression of polyQ-expanded protein aggregation, whereby only a subset of DNAJB isoforms were effective (Hageman et al., 2010). Whilst we identified that the TTK-LKS motif in DNAJB6 and DNAJB8 is essential for these proteins to prevent the intracellular aggregation of destabilised client proteins, such as FlucDM, this does not account for the capacity of other DNAJBs (that lack this region) to suppress inclusion formation by FlucDM. Our data are nevertheless largely consistent with findings on mutant Parkin, in which the overexpression of most DNAJ proteins tested reduced the propensity of the mutant Parkin to form amorphous aggregates (Kakkar et al., 2016a). Furthermore, in the case of an ALS-causing mutant SOD1 protein, overexpression of multiple DNAJs (albeit not all) supressed aggregation (Serlidaki et al., 2020). In all these cases, including our current data, the effects coincided with a reduction in steady-state levels of the mutant protein. Although we cannot formally exclude effects of quality control during co-translational folding, we favour the hypothesis that the various DNAJ proteins recognise and bind to these (partially) misfolded substrates post-translationally to support their proteasomal degradation (Kakkar et al., 2016a; this report). This mechanism is distinct from that seen for polyQ proteins, which aggregate in a precise and ordered manner that can be chaperoned by distinct binding regions present only in the DNAJB6-like proteins (i.e. DNAJB2, DNAJB6, DNAJB7 and DNAJB8). The global unfolding of the other more structurally destabilised proteins may expose many hydrophobic surfaces that can be recognised by the multiple different substrate-binding sites in DNAJB1-like proteins and by other regions of the DNAJB6-like chaperones. Future studies to elucidate the specific region(s) within DNAJB1-like proteins that act to suppress the aggregation of destabilised client proteins could utilise a similar approach to that undertaken in this work, by encompassing a range of deletion mutations located throughout the C-terminal substrate-binding domain(s).
In conclusion, we have utilised the proteostasis sensor FlucDM to demonstrate that overexpression of the DNAJB molecular chaperones acts to boost the PQC capacity of cells. We demonstrate that the ability of DNAJBs to inhibit the aggregation of FlucDM into inclusions relies on interaction with Hsp70, and this facilitates degradation of FlucDM by the proteasome. Significantly, we show that the TTK-LKS region in the C-terminal domain of DNAJB6 and DNAJB8 is essential for engaging this destabilised client protein to prevent its aggregation. Moreover, we show that the S/T-rich region of DNAJB6-like proteins that mediates interactions with amyloid-forming client proteins is not involved in suppressing FlucDM aggregation. Overall, our data emphasises the important role of DNAJB molecular chaperones in preventing all forms of protein aggregation in cells and highlights the potential of targeting them for the amelioration of diseases associated with protein aggregation.
MATERIALS AND METHODS
The enhanced green fluorescent protein (EGFP)-N3 plasmid was donated by Dr Darren Saunders (University of New South Wales, Sydney, NSW, Australia). Plasmids encoding wild-type (WT) and double mutant (DM; R188Q, R261Q) Fluc with an N-terminal EGFP tag (FlucWT–EGFP and FlucDM–EGFP) (Gupta et al., 2011) were kindly gifted by Professor Ulrich Hartl (Max Planck Institute of Biochemistry, Munich, Germany) and were cloned into pcDNA4/TO/myc/hisA for mammalian expression by GenScript (Piscataway, NJ, USA). The construction of the V5-tagged DNAJB plasmid library used in this study is described in Hageman and Kampinga (2009). Plasmids expressing pcDNA5/FRT/TO-monomeric red fluorescent protein (mRFP), mutations in DNAJBs in which a histidine residue is replaced with a glutamine (H/Q) within the J domain and C-terminal deletions in DNAJB8 (ΔSSF-SST and ΔTTK-LKS) are outlined in Hageman et al. (2010). Plasmids encoding mutations in the S/T-rich region of DNAJB6 have previously been described by Kakkar et al. (2016b) and similarly referred to herein as M1–M4. Constructs expressing disease-related missense mutations (F93L and P96R) in the G/F-rich region have previously been described by Thiruvalluvan et al. (2020). The plasmid encoding deletion of the TTK-LKS region in DNAJB6 was cloned by Dr Jurre Hageman (Hanze University of Applied Sciences, Groningen, The Netherlands).
Cell culture, transient transfections and treatment
HEK293 cells (American Type Culture Collection, Manassas, VA, USA) were cultured in DMEM/F-12 supplemented with 2.5 mM L-glutamine (Gibco, Carlsbad, CA, USA) and 10% (v/v) foetal calf serum (FCS; Gibco) at 37°C under 5% CO2 and 95% air in a Heracell 150i CO2 incubator (Thermo Fisher Scientific, Glen Burnie, MD, USA). HEK293 cells stably expressing the tetracycline (tet)-repressor (Flp-In T-REx HEK293; Invitrogen, Carlsbad, CA, USA) were cultured as above with the addition of 50 μg/ml Zeocin and 5 μg/ml blasticidin (Invitrogen) to the culture medium weekly to ensure maintenance of the tet-repressor. Cells were routinely tested for mycoplasma contamination (∼every 6 months) and the identity of these cell lines were verified via short tandem repeat profiling (Garvan Institute of Medical Research, Sydney, NSW, Australia).
For transient transfections, cells were grown to 60–70% confluence in CELLSTAR 6-well plates (Greiner Bio-One, Frickenhausen, Germany) coated with 0.001% poly-L-lysine (Sigma-Aldrich, St Louis, MO, USA). Cells were co-transfected 24 h post-plating with linear (MW 25,000) polyethylenimine (BioScientific, Gymea, NSW, Australia) according to the manufacturer's instructions, and 0.2 µg of plasmid encoding FlucDM and 0.8 µg of plasmid DNA encoding either mRFP (as a negative control) or a DNAJB isoform (wild-type or mutational variant). For transfection in Flp-In T-REx HEK293 cells, 1 µg/ml tetracycline (Sigma-Aldrich) was added to the culture medium 4 h post-transfection to induce expression. For inhibition of the proteasome, 10 µM MG132 (SelleckChem, Boston, MA, USA) was added to cells 24 h post-transfection and cells were incubated for a further 18 or 24 h. Autophagy was inhibited 24 h post-transfection using a combination of 1 µM bafilomycin A1 (Sapphire Bioscience, Redfern, NSW, Australia) and 5 mM 3-methyladenine (AdipoGen, San Diego, CA, USA), and then the cells were incubated for a further 24 h. Because MG132, bafilomycin A1 and 3-methyladenine were dissolved in dimethyl sulfoxide (DMSO; Sigma-Aldrich), an equivalent volume of DMSO was added to control samples. In some experiments, HEK293 cells were grown as above and at 48 h post-plating were heat-shocked at 42°C for 1 h, before being allowed to recover at 37°C for 2 h prior to harvesting.
Inclusions formed following expression of FlucWT–EGFP or FlucDM–EGFP were analysed directly in 6-well plates 48 h post-transfection by epifluorescence microscopy. Fluorescence from EGFP was detected following excitation at 488 nm. All images were taken at 20× magnification using a Leica DMi8 fluorescence microscope (Leica Microsystems, Wetzlar, Germany). Images were prepared with the Leica Application Suite – Advanced Fluorescence (LAS-AF) Version 3 software (Leica Microsystems).
Immunocytochemistry and confocal microscopy
Immunocytochemistry and confocal microscopy was performed to detect the co-expression of FlucDM–EGFP and V5-tagged DNAJBs or mutational variants (or mRFP). HEK293 cells were grown to 60–70% confluency in 8-well chamber µ-Slides (Ibidi, Martinsried, Germany) and transfected as described above but using a tenth of the amount of plasmid DNA and transfection reagent so as to maintain the same ratio of these reagents in the volume of culture medium as for transfections performed in 6-well plates. At 48 h post-transfection, cells were pre-fixed with warmed (37°C) 2% (v/v) paraformaldehyde (PFA) for 5 min, prior to fixing with 4% (v/v) PFA for 15 min at room temperature. Cells were washed twice with 100 mM Tris-HCl (pH 8.0) for 10 min with gentle rocking to quench residual PFA. Cells were then permeabilised with 0.1% (v/v) Triton X-100 (Thermo Fisher Scientific) in phosphate-buffered saline (PBS) for 10 min and blocked for 1 h at room temperature in blocking buffer [1% (v/v) FCS, 1% (w/v) BSA and 0.1% (v/v) Triton X-100 in PBS, pH 7.4]. Cells were incubated with an anti-V5 antibody (1:200; 46-0705, Thermo Fisher Scientific) diluted in blocking buffer for 1 h at 37°C and then washed three times (each for 10 min) with gentle rocking in 0.1% (v/v) Triton X-100 in PBS. Cells were then incubated with the goat anti-mouse IgG H&L Dylight 550 secondary antibody (1:250; ab96872, Abcam, Cambridge, MA, USA) diluted in blocking buffer for 30 min at 37°C in the dark, prior to three washes (each for 10 min) in 0.1% (v/v) Triton X-100 in PBS with rocking. Finally, cells were stained with 0.1 µg/ml Hoechst 33342 nucleic acid stain (Thermo Fisher Scientific) for 5 min at room temperature, washed twice in PBS and imaged in ∼150 µl PBS/well using a SP8 TCS confocal microscope. For imaging, the 63× oil objective lens and, where required, the 4× zoom function in the LAS-X software (Leica Microsystems) was used. To eliminate spectral overlap, fluorescence images were acquired by sequential scanning, where Hoechst 33482 was excited with a 405 nm laser, EGFP was excited with a 488 nm laser and mRFP or Dylight 550 (V5-tag) were excited with a 552 nm laser. Images were prepared using the LAS-X Version 3 software.
Flow cytometry assay to assess inclusion formation
In some experiments, 48 h post-transfection, cells were prepared for flow cytometric analysis using the flow cytometric analysis of inclusions and trafficking (FloIT) method previously described (Whiten et al., 2016). To do so, cells were harvested with 0.05% (v/v) trypsin/EDTA (Gibco), then diluted with DMEM/F-12 containing 1% (v/v) FCS and centrifuged at 300 g for 5 min at room temperature. Cells were then washed twice in PBS (135 mM NaCl, 2.7 mM KCl, 1.75 mM KH2PO4, 10 mM Na2HPO4, pH 7.4) and resuspended in 500 μl PBS. Cells were kept on ice throughout this process to minimise cell death and protein aggregation. An aliquot of the cell suspension (150 μl) was taken and used to measure the transfection efficiency of live cells for use in later analyses.
Flow cytometry was performed using a BD LSRFortessa X-20 or BD LSR-II analytical flow cytometer (BD Biosciences, San Jose, CA, USA), and FCS files were analysed using FlowJo version 10 (Tree Star Ashland, OR, USA). Histograms were generated and statistical analyses were performed using GraphPad Prism version 8 (GraphPad Software, San Diego, CA, USA). Unless otherwise stated, results are reported as the mean±s.e.m., and the number of independent (biological) replicates (n) of each experiment is specified. Data were analysed by one-way ANOVA and Tukey's post-hoc test or, where appropriate, assessed assuming unequal variance using the unpaired, two-tailed Student's t-test. In all analyses, P<0.05 was considered statistically significant.
Cellular protein extraction, quantification and fractionation
Transfected or heat-shocked cells were trypsinised, harvested, washed twice in PBS (300 g for 5 min at room temperature), and total cellular protein was extracted by lysis with Nonidet P-40 (NP-40; Thermo Fisher Scientific) lysis buffer [50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA and 1% (v/v) NP-40 supplemented with 0.5% (v/v) Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific), pH 8.0]. Cell lysates were then sonicated using the Sonifer 250 Digital cell disruptor and a double step micro-tip (Branson Ultrasonics, Brookfield, CT, USA) at 50% amplitude for 5 s. The total protein concentration for each sample was then determined using a BCA assay (Thermo Fisher Scientific) according to the manufacturer's instructions. The concentration in each sample was adjusted with NP-40 lysis buffer to generate cell lysates of 1 mg/ml total protein (total volume was 200 µl) to ensure equal loading onto SDS polyacrylamide gel electrophoresis (SDS–PAGE) gels for subsequent immunoblotting. A 45 µl aliquot of total protein (total fraction) was taken and kept on ice until use. The remaining 155 µl lysate was centrifuged at 20,000 g for 30 min at 4°C and the supernatant (NP-40-soluble fraction) carefully collected and placed on ice. The pellet was washed in ice-cold TNE buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, pH 8.0) and centrifuged again at 20,000 g for 30 min at 4°C. The supernatant was carefully removed and discarded, and the pellet resuspended in 50 µl NP-40 lysis buffer. The insoluble pellet was sonicated at 50% amplitude for 5 s (NP-40 insoluble fraction). SDS–PAGE loading buffer [final concentrations: 500 mM Tris-HCl, 2% (w/v) SDS, 25% (w/v) glycerol, 0.01% (w/v) Bromophenol Blue and 15% (v/v) β-mercaptoethanol (Sigma-Aldrich), pH 6.8] was added to each sample and the samples were then heated at 95°C for 5 min.
Immunoblotting and detection
Equal amounts of protein were loaded onto 12% (v/v) resolving SDS–PAGE gels with 4% (v/v) polyacrylamide stacking gels using Precision Plus Protein dual-colour standards (Bio-Rad, Hercules, CA, USA). The gels were run in a Mini-Protean Tetra Cell system (Bio-Rad) filled with SDS–PAGE running buffer [25 mM Tris base, 192 mM glycine and 0.5% (w/v) SDS, pH 8.3]. Samples were electrophoresed for 15 min at 100 V until proteins had migrated through the stacking gel, at which point the voltage was increased to 150 V and samples were allowed to run until the Bromophenol Blue dye front had migrated beyond the end of the gel (∼1 h). Proteins were transferred onto an ImmunoBlot polyvinylidene difluoride (PVDF; Bio-Rad) membrane at 100 V for 1 h in ice-cold transfer buffer [25 mM Tris base, 192 mM glycine, 20% (v/v) methanol, pH 8.3]. The membrane was blocked at 4°C overnight with 5% (w/v) skim milk powder in Tris-buffered saline (TBS; 50 mM Tris base and 150 mM NaCl, pH 7.6). Membranes were incubated with either rabbit anti-GFP (1:2500; ab290, Abcam), mouse anti-Hsp70 (1:1000; ab47455, Abcam), mouse anti-SQSTM1/p62 (1:2000; ab56416, Abcam), mouse anti-ubiquitin (1:1000; sc-8017, Santa Cruz Biotechnology, Dallas, TX, USA) or mouse anti-V5 (1:5000; 46-0705, Thermo Fisher Scientific) primary antibodies in 5% (w/v) skim milk powder in TBS containing 0.05% (v/v) Tween 20 (TBS-T) for 2 h at room temperature. The membrane was washed four times (each for 10 min) in TBS-T before being incubated with an anti-mouse IgG horseradish peroxidase (HRP)-conjugated secondary antibody (A9044, Sigma-Aldrich) or an anti-rabbit IgG HRP-conjugated secondary antibody (31466, Thermo Fisher Scientific), each diluted 1:5000 into 5% (w/v) skim milk powder in TBS-T. The membrane was rocked at room temperature for 1 h before being washed four times (each for 10 min) in TBS-T. Proteins of interest were detected with SuperSignal West Pico Chemiluminescent Substrate or SuperSignal West Dura Extended Duration Chemiluminescent Substrate (Thermo Fisher Scientific) using an Amersham Imager 600RGB (GE Healthcare Life Sciences, Little Chalfont, UK) or ChemiDoc Imaging System (Bio-Rad), with exposure times ranging from 1 to 15 min.
We thank Ms Maria van Waarde-Verhagen for technical assistance and staff in Molecular Horizons and the Illawarra Health and Medical Research Institute for technical and administrative support.
Conceptualization: S.B., H.H.K., H.E.; Methodology: S.M., S.B., H.H.K., H.E.; Validation: S.M.; Formal analysis: S.M., H.E.; Investigation: S.M.; Resources: S.B., H.H.K., H.E.; Data curation: S.M.; Writing - original draft: S.M.; Writing - review & editing: S.M., S.B., H.H.K., H.E.; Supervision: H.H.K., H.E.; Project administration: H.E.; Funding acquisition: S.B., H.H.K., H.E.
S.M. was supported by a Research Training Program Scholarship from the Department of Education, Skills and Employment, Australian Government and a New Holland Scholarship presented by Nuffic. This work was supported by grants from Ciência sem Fronteiras (Science without Borders) from the Brazilian Government (to H.H.K.), the Dutch Campaign Team Huntington (to S.B. and H.H.K.) and The Netherlands Organization for Scientific Research (ZonMw; project numbers 733051076 and 91217002 to H.H.K.).
Peer review history
The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/134/7/jcs255596/
The authors declare no competing or financial interests.