The prion-like RNA-processing protein HNRPDL forms inherently toxic amyloid-like inclusion bodies in bacteria

HNRPDL displays a predicted amyloidogenic prion-like domain at the C-terminus

The heterogeneous nuclear ribonucleoprotein D-like, also known as HNRPDL, belongs to the subfamily of ubiquitously expressed heterogeneous
nuclear ribonucleoproteins (hnRNPs). These proteins are associated with pre-mRNAs
in the nucleus, functioning in mRNA biogenesis and mRNA metabolism 46]. Although all of the hnRNPs are present in the nucleus, some shuttle between the
nucleus and the cytoplasm 48]. HNRPDL is a 420 residues long protein for which no structural information is available
yet. Both SMART (http://smart.embl-heidelberg.de) and PFAM (pfam.sanger.ac.uk/) databases coincide to indicate the presence of two
contiguous canonical RNA recognition motifs (RRM) including residues 149–221 and 234–306,
occupying a central position in the protein (Figure 1). Both the N- and C- terminal boundaries of these small domains are predicted to
be low complexity regions without any associated function or structural motif. Disorder
predictions using FoldIndex 49], FoldUnfold 50], and RONN 51] algorithms suggest that both the 1–149 and 306–420 sequence stretches are essentially
disordered (Figure 1). The amino acid compositional bias of Q/N enriched prion domains has allowed the
recent development of three different algorithms to identify the presence of PrLDs
in protein sequences: PAPA 52], PLAAC 53] and PrionScan 54]. No prionic propensity is predicted with any of these programs for the N-terminal
segment, whereas all of them identify the C-terminal region as displaying a PrLD comprising
residues 340–420. Overall, this domain architecture and PrLD location recapitulates
that of TDP-43 (Figure 1; Table 1).

Figure 1. TDP-43 and HNRPDL domain architecture. Cartoons of proteins TDP-43 and HNRPDL show
the domain architecture, where RRM accounts for RNA recognition motif and are represented
in blue, and predicted disordered regions and prion domains (PrD) are shown in striped green and red, respectively. The places where RRM domains as assigned according to PFAM overlay
with disordered predicted regions were assumed to correspond to canonical RRM domains.

Table 1. Prediction of PRLDs and their amyloid cores potency in the sequences of HNRPDL and
TDP-43 RNA-binding proteins

We have recently shown that the identification and evaluation of the potency of amyloid
nucleating sequences in the context of disordered Q/N rich protein segments allows
discrimination of genuine yeast prions from non-prionic sequences displaying very
similar amino acid composition, a concept that was implemented in the pWALTZ algorithm
34]. The C-terminal PrLD of HNRPDL displays a pWALTZ score (82.27) higher than the corresponding
PrLD in TDP-43 (68.16) (Table 1) and, strikingly, higher than those of Ure2p (73.99) and Sup35 (73.66) prion domains
34], thus indicating the presence of an amyloidogenic sequence stretch comprising residues
342–362 in this Q/N rich disordered protein region.

Aggregation of HNRPDL into IBs in bacteria

The inherent aggregation propensity of human amyloid proteins results in most of them
aggregating into insoluble IBs when they are produced in bacteria 55]. To test if this is the case of HNRPDL, we analyzed the cellular distribution of
the recombinant protein after its expression in E. coli at 37°C for 20 h. As assessed by SDS-PAGE, a new protein band of ~50 kDa, corresponding
to the expected HNRPDL molecular weight (47 kDa), could be detected in induced cells
(Figure 2a). The bacteria cells were harvested, lysed and centrifuged and the resulting supernatant
and pellet fractions were analyzed by SDS-PAGE. HNRPDL was found essentially in the
insoluble fraction suggesting that it likely aggregated into IBs (Figure 2a). The protein remained in the insoluble fraction when protein expression was induced
at either 25 or 18°C (data not shown). We further cloned the HNRPDL cDNA downstream
of the GST gene in a pETM30 vector and expressed the fusion protein at 20°C for 20 h.
A new protein band of ~75 kDa was observed for induced cells, corresponding to the
sum of the molecular weights of GST (26 kDa) and HNRPDL (47 kDa) (Figure 2b). Fractionation indicated that despite the theoretical solubility provided by GST,
the fusion was located in the insoluble fraction (Figure 2b) a localization that was maintained when protein expression experiments were performed
at lower temperatures (data not shown). Because RRM domains are known to be soluble
at high concentrations 56] and no aggregation-prone region is detected at the disordered N-terminal segment
using predictive algorithms like AGGRESCAN 57] or TANGO 58], it is likely that the predicted amyloidogenicity of the prion-like C-terminal region
would account for the propensity of HNRPDL to form intracellular aggregates, either
alone or when fused to GST.

Figure 2. Expression of recombinant HNRPDL protein in E. coli cells. a Analysis on SDS-PAGE of E. coli cells extracts expressing HNRPDL protein. b SDS-PAGE analysis of cell extracts from cells expressing the GST-HNRPDL fusion. On
both gels lane 1 shows total extract; lane 2, soluble fraction (supernatant), and lane 3 insoluble fraction (pellet). Arrows indicate the bands corresponding to HNRDPL protein.

HNRPDL IBs bind to thioflavin-S in living cells

We have shown recently that thioflavin-S (Th-S) staining of living bacterial cells
can be used to detect the presence of intracellular amyloid-like structures as well
as to find inhibitors that interfere with amyloid formation 17], 59]. The staining of cells expressing HNRPDL was monitored using confocal microscopy.
As it can be observed in Figure 3a, induced cells exhibited a green fluorescent background with strong fluorescent
foci located at the cell poles, suggesting that HNRPDL adopts amyloid-like conformations
in bacterial IBs. In contrast, non-induced control cells exhibited only residual fluorescence.
The presence of intracellular amyloid-like protein conformations in induced cells
could also be monitored using fluorescence spectroscopy. As previously described for
cells expressing A?42 59], the Th-S fluorescence maximum increases and red-shifts in the presence of living
cells expressing HNRPDL, relative to the Th-S fluorescence maximum recorded in the
presence of non-induced cells (Figure 3b).

Figure 3. Th-S staining of cells expressing HNRPDL. a Fluorescent confocal microscopy images of non-induced E. coli cells and expressing HNRPDL IBs stained with Th-S at ×100 magnification. b Fluorescence spectra of Th-S in the presence of non-induced (?IPTG) and induced (+IPTG)
living cells expressing HNRPDL. Arrows indicate the position of IBs.

Purified HNRPDL IBs bind to amyloid dyes

We next purified the HNRPDL IBs to characterize biophysically their amyloidogenic
properties. Using SDS-PAGE densitometry we calculated that HNRPDL constituted around
30% of all proteins in the purified IBs fraction (Figure 4). To evaluate the specific contribution of HNRPDL in the different assays, relative
to that of other proteins present in this fraction, cells bearing the same plasmid
without any insert were induced and the IBs fraction purified in the same manner than
those containing the HNRPDL cDNA and used as negative control (Figure 4). In addition, the IBs of cells expressing the yeast prion Ure2p and A?42 were purified
using the same protocol and used as positive controls, since extensive characterization
of the bacterial IBs formed by these two proteins have revealed that they posses an
amyloid-like nature 16], 27].

Figure 4. Purification of recombinant HNRPDL IBs. SDS-PAGE analysis of IBs purified from the
insoluble fraction of induced cells grown at 37°C containing either an empty plasmid
(lane 1) or a plasmid encoding for HRNPDL (lane 2). The arrow indicates the band corresponding to HNRDPL.

Thioflavin-T (Th-T) fluorescence emission is enhanced in the presence of amyloid fibrils
60]. Consistent with their amyloid properties, the same behaviour is observed upon incubation
of Th-T with A?42 and Ure2p IBs. In the same way, the increase in Th-T fluorescence
in the presence of HNRPDL IBs suggests the existence of amyloid conformations in the
polypeptides embedded in these aggregates (Figure 5a). Although their impact in Th-T fluorescence is lower than that of A?42 IBs, it
is quite similar to the one promoted by Ure2p IBs and remarkably different from that
observed in the IBs fraction of negative control cells.

Figure 5. Specific binding of amyloid dyes to HNRPDL IBs. a Fluorescence emission spectra of Th-T in the absence or the presence of A?42, Ure2p,
HNRPDL and control IBs. b Congo red (CR) absorbance spectra in the absence or the presence of A?42, Ure2p,
HNRPDL and control IBs. c Difference absorbance spectra of CR in the presence and in the absence of IBs, showing
the characteristic amyloid maximum at 540 nm.

The absorbance of the amyloid dye congo red (CR) red-shifts in the presence of amyloid
fibrils 61]. The same effect was observed in the presence of A?42, Ure2p and HNRPDL IBs, consistent
with the presence of amyloid-like structures in these aggregates. The observed red-shift
was smaller for HNRPDL than for the other two amyloid proteins, but still significantly
different from that promoted by the IBs fraction of negative control cells (Figure 5b). Indeed, quantification of CR bound to IBs (see “Methods”) indicates that HNRPDL
binds 2.4 times more dye than control IBs. The difference spectrum between the dye
in the absence and presence of purified IBs allows the detection of the characteristic
band at 540 nm, corresponding to the amyloid conformation in the three IBs (Figure 5c).

HNRPDL IBs are enriched in intermolecular ?-sheet structure

From a structural point of view, the formation of amyloid fibrils is always characterized
by an enrichement in protein ?-sheet content 61]. Attenuated Total Reflectance–Fourier Transform Infrared spectroscopy (ATR-FTIR)
is a powerful tool to investigate the secondary structure in protein aggregates 62]–65]. We used this technique to analyse the conformational properties of the IBs in the
present study (Figure 6; Table 2; Additional file 1: Figure S1). Deconvolution of the absorbance spectra in the amide I region allows
to observe a signal at ~1,622 cm
^?1
common to the IBs formed by A?42, Ure2p and HNRPDL proteins, which is otherwise absent
in negative control samples. This band is usually attributed to the presence of densely
packed ?-sheet structures, linked by short and strong hydrogen bonds, compatible with
the intermolecular contacts in an amyloid fold 62]. A?42, Ure2p and HNRPDL IBs also share a band at ~1,636 cm
^?1
, which has been typically assigned to intramolecular ?-sheet; this band is also present
in the negative control, but it contributes less to the total spectral area. In contrast,
the negative control IBs exhibits higher contributions at ~1,653 cm
^?1
and ~1,665 cm
^?1
, which indicates an enrichment in helical, irregular and turn conformations, relative
to A?42, Ure2p and HNRPDL IBs. A?42 and Ure2p IBs display a band at 1,682 cm
^?1
, which is usually assigned to a high frequency ?-sheet signal 66]. The lack of this signal, together with the presence of an exclusive band at ~1,676 cm
^?1
, attributed to turns 66], suggests that despite sharing an amyloid nature, the fine structural properties
of HNRPDL IBs differ from those formed by A?42 and Ure2p.

Figure 6. Secondary structure content of HNRPDL IBs. FTIR absorbance in the amide I region of
the infrared spectrum (black) for A?42, Ure2p, HNRPDL and control IBs. Spectral components in the Fourier deconvoluted
FTIR spectra are shown. The area and position of the correspondent bands are indicated
in Table 2.

Table 2. Contribution of secondary structure components to the absorbance FTIR spectra of A?42,
Ure2p, HNRPDL and control IBs

HNRPDL IBs posses an inner amyloid core

We monitored the morphology of HNRPDL IBs using Transmission Electronic Microscopy
(TEM). Freshly purified IBs displayed a typical electrodense amorphous appearance
(Figure 7). However, upon incubation of purified IBs at 37°C for 12 h, the presence of fibrillar
structures becomes already evident (Figure 7). The same behaviour has been reported for the amyloid-like IBs of other proteins
and interpreted as the IBs containing densely packed bundles of amyloid fibrils inside
cells that become relaxed and exposed upon in vitro incubation 14]. This property can be qualitatively tested using proteinase K (PK), a protease usually
used to map the protected core of amyloid fibrils because in spite of being highly
active against peptidic bonds it cannot easily attack the highly packed backbones
in amyloid ?-sheet structures. Accordingly, we have shown that PK digestion allows
revealing the existence of a fibrillar core in A? peptide IBs 15]. We used the same approach to assess if the presence of a similar fibrillar material
might account for the amyloid conformational properties of HNRPDL IBs. Upon PK digestion,
the presence of typical long and unbranched amyloid fibrils becomes evident. The fibrils
are associated with apparently amorphous material and in some micrographs fibrils
emerging from the preformed compact IBs are seen. The elementary fibrils are ~5 nm
in diameter and tend to associate laterally into bundles, thus supporting that HNRPDL
IBs constitute a bacterial reservoir of amyloid structures, that coexist with less
ordered and PK susceptible protein regions, in good agreement with the deduced secondary
structure content from FTIR analysis. According to the presence of an amyloid core:
(1) HNRPDL IBs are much more resistant towards PK digestion than negative control
IBs (Additional file 2: Figure S2) and (2) HNRPDL IBs retain significantly higher Th-T binding in diluted
solutions than negative control IBs even upon long time incubation (Additional file
3: Figure S3). These two properties recapitulate that of the amyloid-like IBs formed
by A?40 and A?42 peptides in bacteria 16].

Figure 7. HNRPDL IBs contain amyloid-like fibrils. Negatively stained HNRPDL IBs visualized
by TEM. The upper panel shows freshly purified HNRPDL IBs (left) and IBs incubated overnight at 37°C (right). The bottom panel displays representative micrographs of PK digested HNRPDL IBs.

HNRPDL IBs are toxic to cultured neuronal cells

It has been shown for different and unrelated proteins that the binding to ANS-like
dyes correlates with the toxicity of amyloid species, suggesting that the exposure
of hydrophobic patches is a critical characteristic of these pathogenic assemblies
67]. We analyzed the binding of bis-ANS to A?42, Ure2p and HNRPDL IBs. In the presence
of these aggregates, bis-ANS experienced the expected blue-shift and a strong increase
in fluorescence maximum. The strongest spectral changes were promoted by the A?42,
and the prion Ure2p IBs. However, HNRPDL IBs induced a significantly higher increase
in bis-ANS fluorescence than negative control IBs (Figure 8).

Figure 8. Binding of bis-ANS to HNRPDL IBs. Fluorescence spectra of bis-ANS in the absence and
presence of A?42, Ure2p, HNRPDL and control IBs.

The aggregates formed by different human prion-like proteins have been shown to exert
neurotoxicity 68]; therefore we tested if, in agreement with their bis-ANS binding ability, purified
HNRPDL IBs could be toxic for cultured neuroblastoma SH-SY5Y cells. The combination
of Hoechst and propidium iodide (PI) staining allows to asses cell viability by fluorescence
microscopy, as viable cells are permeable to Hoechst and PI only enters cells with
disintegrated membranes thus corresponding to dead cells. Cell morphology can be monitored
as well to discriminate toxic and non-toxic aggregates in this assay. In samples treated
with negative control IBs cell were attached to the culture plate at a confluent stage
with only a reduced number of cells becoming stained with PI, indicating that they
display low or no toxicity (Figure 9). In contrast, the IBs formed by A?42 and Ure2p proteins were inherently toxic to
neuronal cells as both induce positive PI staining in most cell nuclei (Figure 9). In the same manner, HNRPDL IBs turned to be highly neurotoxic, with a large majority
of cells being stained by PI (Figure 9). Moreover, this effect was dose dependent, since cells incubated with 40 µg/mL of
HNRPDL IBs kept attached, homogeneously distributed and displayed normal morphology,
whereas cells treated with 80 µg/mL HNRPDL IBs lost completely their morphology becoming
detached and agglutinated (Figure 9).

Figure 9. Toxicity of HNRPDL IBs as visualized by confocal microscopy. Representative confocal
fluorescence microscopy images of SH-SY5Y cells stained with propidium iodide (IP)
or Hoechst after incubation with A?42, Ure2p, HNRPDL and control IBs for 24 h at 37°C.
The bar corresponds to 15 µM.