Thyroid hormone synthesis continues despite biallelic thyroglobulin mutation with cell death

Complete absence of thyroid hormone is incompatible with life in vertebrates. Thyroxine is synthesized within thyroid follicles upon iodination of thyroglobulin conveyed from the endoplasmic reticulum (ER), via the Golgi complex, to the extracellular follicular lumen. In congenital hypothyroidism from biallelic thyroglobulin mutation, thyroglobulin is misfolded and cannot advance from the ER, eliminating its secretion and triggering ER stress. Nevertheless, untreated patients somehow continue to synthesize sufficient thyroxine to yield measurable serum levels that sustain life. Here, we demonstrate that TGW2346R/W2346R humans, TGcog/cog mice, and TGrdw/rdw rats exhibited no detectable ER export of thyroglobulin, accompanied by severe thyroidal ER stress and thyroid cell death. Nevertheless, thyroxine was synthesized, and brief treatment of TGrdw/rdw rats with antithyroid drug was lethal to the animals. When untreated, remarkably, thyroxine was synthesized on the mutant thyroglobulin protein, delivered via dead thyrocytes that decompose within the follicle lumen, where they were iodinated and cannibalized by surrounding live thyrocytes. As the animals continued to grow goiters, circulating thyroxine increased. However, when TGrdw/rdw rats age, they cannot sustain goiter growth that provided the dying cells needed for ongoing thyroxine synthesis, resulting in profound hypothyroidism. These results establish a disease mechanism wherein dead thyrocytes support organismal survival.


Introduction
In the body, the circulating thyroid hormone, thyroxine (also known as T 4 ) originates exclusively from biosynthesis within the thyroid gland. Thyroxine biosynthesis occurs by a common mechanism in all vertebrates. Specifically, a monolayer of thyrocytes (also known as thyroid follicular epithelial cells) surrounds a central apical (extracellular) lumen, into which thyrocytes deliver a nearly-pure secretion of thyroglobulin (Tg, encoded by the TG gene) (1), which comprises ≥ 50% of the total protein of the thyroid gland (2). Thyrocytes exhibit a polarized distribution of plasma membrane enzymes/activities that coordinate thyroid peroxidase-catalyzed apical iodination of extracellular protein in the luminal cavity (3). Iodination of various tyrosine residues on secreted Tg (4) triggers the formation of T 4 intramolecularly within the Tg protein (5,6) prior to endocytic re-entry of the hormone-containing protein into surrounding thyrocytes for lysosomal digestion, resulting in the proteolytic liberation and release of T 4 from the basolateral membrane of thyrocytes to the bloodstream (7).
The first three-dimensional atomic structure of human Tg has recently been reported (8).
Already 227 different TG gene mutations have been found to be linked to congenital hypothyroidism (9); as far as is known, essentially all of the structurally-defective Tg mutants are entrapped in the endoplasmic reticulum (ER), causing thyrocyte ER swelling and ER stress (10). Susceptibility to the many different pathogenic mutations is in part explained by the large and complex structure of the Tg protein (8), including its multiple repeat domains bearing internal disulfide bonds, and concluding with the Cholinesterase-Like (ChEL) domain (10). The C-terminal ChEL domain of Tg has no direct impact on thyroidal iodination machinery, but it: a) shares a similar structure with other ChEL family members (11), b) provides information necessary and sufficient for the noncovalent homodimerization needed for intracellular transport (12), c) functions as an intramolecular chaperone required to stabilize the folded structure of upstream repeat domains of Tg (13), and d) provides its own hormonogenic iodination site (14). A number of human patients have been reported with homozygous mutation in the Tg-ChEL domain (e.g., Tg-W2346R or Tg-G2322S; in the UNIPROT P01266 numbering system this would need to include the 19-residue signal peptide) with congenital hypothyroidism (15,16).
In years past, cases of congenital hypothyroidism could go undiagnosed in early life due to insufficient neonatal screening (17). Classic studies of Marine and Lenhart showed that thyroid hyperplasia is induced as a consequence of primary hypothyroidism (experimentally-induced following partial thyroidectomy or iodide deficiency) in animals (18) or humans (19); i.e., the endocrine feedback of primary hypothyroidism results in chronically upregulated pituitary secretion of Thyroid Stimulating Hormone (TSH) and such chronic stimulation contributes to exuberant growth of the thyroid gland (20)(21)(22). Therefore, patients with bi-allelic TG mutations would be expected to present, ultimately, with goiter. Interestingly, however, by linkage analysis, variants of the TG gene are linked to human hypothyroidism with or without thyroid goiter (23). Why hypothyroid patients with bi-allelic TG mutation (and no defect in TSH response) would not develop a goiter is unknown, although in the clinical setting, the understanding of goiter development is often confounded in patients who may or may not have received exogenous thyroxine treatment (24)(25)(26).
On the one hand, increased goiter growth might help to overcome genetic or acquired inefficiency of thyroid hormone synthesis (27); on the other, growth of a large goiter in iodine deficiency has been proposed to be a maladaptation (28). In either case, the fundamental knowledge gap has been an understanding of how untreated patients bearing pathogenic, bi-allelic TG mutations could possibly be capable of synthesizing endogenous T 4 .
Chronic ER stress is known to occur in the thyrocytes of TG cog/cog (congenital goiter) mice (encoding Tg-L2263P) which are famous for their hyperplastic goiter -and also in TG rdw/rdw rats (35) that do not develop a goiter (36). In all cases of bi-allelic TG mutation, it is thought that massive quantities of mutant Tg protein are blocked in forward advance from the ER, as in the TG cog/cog and TG rdw/rdw thyroid glands, triggering a dramatic ER stress response that is also seen in the thyroid glands of human patients with this disease (35,(37)(38)(39). Thyrocyte cell death has never been considered in goitrous TG cog/cog mice or humans with bi-allelic TG mutations, but in TG rdw/rdw rats we posited that thyroid follicular cell death might block the development of goiter (40). Importantly, untreated TG cog/cog mice spontaneously increase their levels of serum T 4 that parallels growth of the thyroid gland, ultimately achieving nearly-normal levels (41). With this in mind, in this report we have analyzed both human and rodent thyroid glands of individuals expressing biallelic TG missense mutations that render Tg incapable of forward trafficking from the ER.
Remarkably, we find that thyrocyte cell death and disintegration within the thyroid follicle lumen provides the Tg substrate needed for synthesis of endogenous T 4 . The life of untreated individuals depend on this unusual mechanism of endogenous T 4 synthesis, as even a brief exposure of such animals to antithyroid drugs is lethal. Most remarkably, we have uncovered compelling evidence that in this disease, goiter growth is needed to provide an ongoing supply of dead cells in order that thyroid hormonogenesis can be sustained.  (46,47), the foregoing data do suggest that in the TG cog/cog thyroid gland, Tg must arrive in the lumen of thyroid follicles via a delivery mechanism other than the conventional secretory pathway. Interestingly, the patchy distribution of mutant Tg in the follicle lumen appeared associated with additional cellular material, including nuclear chromatin (Fig. 1A, B).
These data (and additional evidence, below) led us to consider that T 4 synthesis in TG cog/cog mice might be based on mutant Tg being delivered to the thyroid follicle lumen via thyrocyte cell death. Some of the T 4 synthesized in Tg can occur within small peptide regions that do not require the native globular structure of the entire molecule (8,59), and in pilot studies using a recently-developed assay for thyroid hormone formation after in vitro iodination (14) of transfected cell lysates, we observed that ER-entrapped recombinant mutant cogTg and rdwTg (described below) have the potential to serve as substrate for T 4 synthesis. Immunoblotting of unpurified thyroid homogenates with anti-T 4 to identify T 4 -containing proteins revealed the major Tg hormone-containing fragment [~250 kD (38)] and its degradation products (7) in WT mouse thyroid tissue, whereas TG cog/cog thyroid glands did not immediately reveal a clear predominant species (Supplemental Fig. S1C lanes 2-4).
We selectively concentrated T 4 -containing protein from TG cog/cog thyroid tissue by immunoprecipitation with anti-T 4 , followed by immunoblotting of the recovered samples with a mAb that specifically favors recognition of intact Tg (epitope located between Tg residues 1000-1100). As expected, when no tissue sample was included in the anti-T 4 immunoprecipitation, no T 4 -containing Tg protein was In hypothyroidism with bi-allelic mutant TG, thyroid cell mass is the critical factor regulating thyroid hormone synthesis. The adult homozygous TG rdw/rdw rat (encoding Tg-G2298R) is well-known for congenital hypothyroidism, although rather than goiter, the animal develops a hypoplastic thyroid gland (36,(60)(61)(62). As in other vertebrates, the normal rat thyroid exhibits a classic monolayer of epithelial thyrocytes surrounding a central cavity filled with secreted eosinophilic Tg (Fig. 3A left).
TG rdw/rdw rat thyroid glands also form follicles surrounding a central cavity with eosinophilic content (Fig.   3A right). Although suitable antibodies were not available to confirm the apical distribution of (rat) thyroid peroxidase and DUOX2 (two enzymes that help to catalyze T 4 synthesis), we could confirm that aminopeptidase-N -also known to be an apical membrane marker in thyrocytes (63,64) was still delivered to its correct destination in TG rdw/rdw thyroid follicles (Fig. 3B, a dashed yellow line highlights the basal membrane outlining the outer boundary of thyroid follicles). Nevertheless, the TG rdw/rdw thyroid histology was far from normal -the cytoplasm was massively engorged with eosinophilic "vacuoles" displacing nuclei under the apical plasmalemma, and the staining of the follicle lumen was abnormally heterogeneous (Fig. 3A right). The eosinophilic "vacuoles" are in fact ER (60) filled with the ER molecular chaperone, BiP (Fig. 3C). Whereas >85% of Tg molecules in WT rats was endoglycosidase H-resistant, analysis of TG rdw/rdw thyroid glands (n = 4) showed that the fraction of Tg molecules bearing Endo H-resistance was zero, indicating an inability of mutant Tg to undergo intracellular transport to the Golgi complex (e.g., Fig. 3D), as previously reported (65) We examined ER stress responses in TG rdw/rdw thyroid glands. PERK phosphorylation of eIF2α stimulates increased translation of ATF4 that upregulates CHOP, which (as noted above) has been strongly implicated in cell death (29). In addition to a dramatic increase of BiP and p58ipk, TG rdw/rdw thyroid glands were observed to have increased phosphorylated eIF2α (Fig. 5A), accompanied by a > 10-fold increase of CHOP mRNA (Fig. 5B). A second ER stress-related death pathway involves IRE1 hyperactivation that can trigger a 'terminal UPR' from exuberant RNAse activity ("RIDD"), which typically develops only in cells exhibiting demonstrably high levels of stress-induced IRE1 splicing of XBP1 mRNA (66). We observed that roughly half of thyroidal XBP1 mRNA was spliced to the active form in TG rdw/rdw animals (Fig. 5C), which is impressive considering that thyrocytes and "C-cells" together comprise only ~60% of resident cells in the mouse thyroid (67) We performed Western blotting of total thyroidal proteins with anti-T 4 . Despite the presence of background bands, a ~330 kDa band co-migrating with WT Tg was the clearest T 4 -containing protein specifically identified in the thyroid tissue of untreated TG rdw/rdw rats, and the intensity of this band was completely eliminated by the addition of free T 4 competitor to the antibody incubation during Western blotting (Fig. 5G left). The efficiency of T 4 -formation in this protein indicative of mutant Tg was much less than in the Tg protein from WT rat thyroid glands, especially when considering that more Tg protein was loaded for the mutant sample (Fig. 5G right). Altogether, the data in Figs. 3-5 support that TG rdw/rdw rats also use dead thyrocytes for endogenous T 4 synthesis on mutant Tg.
To more clearly examine the disintegration of dead thyrocytes bearing mutant Tg, fixed/postfixed WT and TG rdw/rdw thyroid tissue were plastic-embedded for semi-thin sectioning. As expected, WT thyroid revealed dense, uniformly-stained "colloid" (Tg protein) in the follicle lumen (Fig. 6A). In contrast, in the thyroid of TG rdw/rdw rats, in addition to large 'vacuoles' in the basal cytoplasm with apically-displaced nuclei, the lumen of different follicles varied, with contents ranging from whole cells to cellular debris (Fig. 6B). Moreover, electron microscopy revealed that living follicular thyrocytes had massively swollen ER with unusual nuclear morphology, and most remaining organelles were crowded into the apical cytoplasm -ultimately limited by the apical plasma membrane bearing microvilli that extend into the follicle lumen (Fig. 6C). Cell ghosts with disintegrating organelles were readily apparent in many of the follicle lumina examined (e.g., Fig. 6D A great puzzle in the field has been to understand why some human patients (and some animal models) with bi-allelic TG mutations that grow a large goiter can yield a survivable serum T 4 level without treatment, yet other human patients and animals models with an intact hypothalamicpituitary-thyroid axis are unable to do so (68). With this question specifically in mind, we examined thyrocyte proliferation in hypothyroid TG rdw/rdw rats. Indeed, in early life we observed that TG rdw/rdw rats did indeed exhibit active proliferation of thyrocytes, similar to that observed in TG cog/cog mice (Supplemental Fig. S4A). Indeed, although never previously described, we observed that in early life TG rdw/rdw rats do in fact develop thyroid enlargement (i.e., goiter) by 9 weeks of age (Fig. 6I) and this parallels a significant increase of endogenous T 4 synthesis that supports serum T 4 levels (Fig. 6J).
However, as the TG rdw/rdw animals aged, the enlarged thyroid gland size could not be sustained (Fig.   6I) and with this (61), the animals could not maintain their serum T 4 levels (Fig. 6J). Untreated profound hypothyroidism is ultimately incompatible with life in rodents [noted above, and (69, 70)] as well as in humans. It thus appears that only patients and animal models that can support a sufficient goiter are able to provide the continuous supply of dead thyrocytes needed for ongoing T 4 synthesis -a mechanism that can allow some individuals the chance to sustain endogenous thyroid hormone levels in adulthood (24).

Discussion
Reports describe untreated adult patients with a large goiter who are biochemically and clinically nearly-euthyroid despite bi-allelic TG deficiency (71)(72)(73). Two longstanding but competing schools of thought are that a) a large hyperplastic goiter is a compensatory physiological adaptation in response to thyroidal genetic or environmental factors that disfavor thyroid hormone production (74), or b) growth of a large thyroid goiter is actually a maladaptation (28). On the one hand, because the Tg protein is the evolutionarily-preferred thyroid hormone precursor (1) Crucially in this disease, it is the dead, disintegrating thyrocytes upon which T 4 -containing protein can be detected, with the thyroid follicles cannibalizing (internalizing) the iodinated detritus of dead thyrocytes into the surrounding living follicular cells.
The spectrum of proteins upon which T 4 might be made inefficiently in humans with bi-allelic TG mutation has not yet been fully explored, although it has often been speculated that albumin, which can become highly iodinated (75) as a serum protein capable of transcytosis (76) or paracelllular leakage (75) -could perhaps be a source of endogenous T 4 . Given the mechanistic understanding presented in the current study, we recognize that the entire proteome leaked from dead thyrocytes becomes eventually exposed to the iodination environment. Here we show that even though the mutant Tg protein cannot be secreted via conventional intracellular trafficking (77), it is nevertheless conveyed to the lumen of thyroid follicles via dead thyrocytes, wherein T 4 is produced (albeit inefficiently) within the mutant Tg protein in TG cog/cog mice, TG rdw/rdw rats and, most likely, humans with the same disease.
Indeed, this pathological salvage mechanism of T 4 synthesis is observed in the goitrous thyroid gland of a patient with homozygous expression of Tg-W2346R. Moreover, in this study, we provide strong supporting evidence that total thyroid cell mass (i.e., the goiter) is important for the endogenous rescue from hypothyroidism. Specifically, increasing amounts of T 4 are produced as the thyroid begins to grow postnatally (41); however in TG rdw/rdw rats, profound hypothyroidism ensues in parallel with an atrophic thyroid gland (36,61). Thus, our results demonstrate that TG rdw/rdw rats form a goiter but cannot sustain their goiter with aging. Evidently, without the growing goiter, the thyroid gland cannot provide sufficient dead cells needed to fuel ongoing thyroid hormone production. These considerations make clear that the balance of cell proliferation-versus-death is indeed a critical factor, as ultimately an unfavorable balance in TG rdw/rdw rats deprives the thyroid of sufficient substrate to maintain T 4 synthesis. In contrast, continued growth of the goiter in TG cog/cog mice allows this pathological salvage mechanism to endogenously self-correct the hypothyroidism (41), and human studies suggest a similar conclusion in goitrous patients with bi-allelic TG mutations.
More work is still needed to determine if rdwTg might somehow be more proteotoxic than cogTg (40) or if other factors that vary between species drive the enhanced capability of TG cog/cog mice for a net proliferation of thyrocytes (i.e., in excess of cell death) into adulthood. What is apparent in all cases, however, is that the continuous contribution of dead and dying thyrocytes to provide substrate for T 4 production represents the critical compensatory response to congenital hypothyroidism with biallelic TG mutations. Moreover, as a brief exposure to antithyroid drug is lethal to TG rdw/rdw rats, our findings appear consistent with the hypothesis that in the presence of bi-allelic TG mutations, survival of the organism (8,68) does require this most unusual means of thyroid hormonogenesis.  Thyroid gland size measurement. Thyroids of euthanized animals were dissected with both lobes of the gland fully exposed. Images of the neck were captured with a calibrated size marker included in situ. The areas of the thyroid glands (correlated with volume) were measured using ImageJ and quantified as a fraction of body weight of each animal.

Primary
Serum total T 4 measurement. Whole blood was collected, clotted and centrifuged at 750 × g for 20 min to obtain serum. Total T 4 was assayed by ELISA (Diagnostic Automation / Cortez Diagnostics).
Preparation and immunostaining of thyroid sections. Thyroid glands from mice and rats were immersion-fixed with 10% formalin and processed for paraffin embedding, sectioning, and H&E staining. For immunofluorescence, 6-µm sections were deparaffinized in Citrisolv and an ethanol series, then heated in citrate buffer (12.3 mM, pH 6) for antigen retrieval, and blocked in 1.5% normal goat serum for 30 min at room temperature. Primary antibody incubation was performed overnight at 4°C, followed by incubation of AlexaFluor-conjugated secondary antibodies (Thermo Fisher). After washing, sections were counterstained and mounted with Prolong-Gold and DAPI (Invitrogen).
Images were captured in a Nikon A1 confocal microscope. For anti-Ki67 immunohistochemistry, the VECTASTAIN ABC Kit (Vector) was used. After antigen retrieval, sections were treated with 3% H 2 O 2 , blocked in 1.5% normal goat serum for 20 min at room temperature, incubated with anti-Ki67 antibody for 1 h at room temperature and biotinylated secondary antibody for 30 min, followed by incubation with avidin-HRP. Staining was visualized by DAB reaction. Sections were counterstained with hematoxylin, dehydrated in a graded series of ethanol, and mounted with Permount. Images were obtained with a Leica DMI-3000B microscope.

T 4 immunofluorescence / TUNEL double labeling. The ApopTag In Situ Apoptosis Detection
Kit (Millipore) was used for TUNEL staining of thyroid sections. Minor modifications were applied for double immunofluorescence labeling with anti-T 4 . Briefly, de-paraffinized thyroid tissue sections were pre-treated with proteinase K (20 µg/mL) and blocked in 1.5% normal goat serum for 30 min at room temperature. Incubation with anti-T 4 antibody was performed at room temperature for 1 h, followed by incubation with AlexaFluor-488-conjugated secondary antibody for 30 min at room temperature. After washing, TUNEL staining was performed. Sections were counterstained and mounted with Prolong-Gold and DAPI (Invitrogen). Fluorescence images were captured in a Nikon A1 confocal microscope.
Endoglycosidase H digest. Thyroid homogenates from WT or mutant mice and rats were boiled in denaturing buffer containing 0.5% SDS and 40 mM DTT at 95°C for 5 minutes, cooled, and then either mock-digested or digested with endoglycosidase H (1000 units, NEB) for 1 h at 37 °C.

Immunoprecipitation analysis of T 4 -containing protein.
Mouse thyroid glands were homogenized in RIPA buffer plus protease inhibitor cocktail (Roche). Thyroid homogenates were incubated with mAb anti-T 4 antibody and protein G-agarose (Exalpha Biologicals) overnight at 4 °C.
Precipitates were washed three times in RIPA buffer (and for samples to be digested with Endo H, two additional washes in PBS) and then boiled in SDS gel-sample buffer containing 50 mM dithiothreitol, resolved by SDS-straight 4.5% or 4-12%-PAGE, electrotransferred to nitrocellulose, and immunoblotted with anti-T 4 or anti-Tg antibody.
PCR. Total RNA was purified from the thyroid gland tissue or PCCl3 cells using a Rneasy Plus kit (Qiagen). Synthesis of cDNA was performed using SuperScript™ III First-Strand Synthesis SuperMix (Invitrogen) or High-Capacity cDNA Reverse Transcription Kits (Thermo Fisher Scientific).
For XBP1 splicing analysis, the primers below were designed to encompass the IRE1 cleavage site of   Representative immunofluorescence of T 4 -containing protein (green) in the thyroid follicle lumen of WT (+/+) and TG rdw/rdw rats (n=5 animals per group). Thyrocyte identity is confirmed by PAX8-positive nuclei (red) with DAPI counter-stain (blue); scale bars = 20 µm. B. Representative immunofluorescence detection of T 4 -containing protein (green; with DAPI counter-stain in blue) in the thyroid of WT and TG rdw/rdw rats (n=9 animals per group) is specifically blocked by addition of T 4 competitor (1 µg/mL). For clarity, a dashed white line delimits the thyroid follicle lumen; a yellow dotted line highlights the outer boundary of the thyroid follicle; scale bars = 10 µm.  5. ER stress, cell death, and T 4 synthesis in TG rdw/rdw rats. A. Left: BiP, p58ipk, and phospho-eIF2α Western blotting in thyroids of WT (+/+) and TG rdw/rdw rats (each lane =1 animal). Right: Quantification (BiP and p58ipk normalized to tubulin; phospho-eIF2α normalized to total eIF2α; mean ± S.D.) ** p <0.01, *** p <0.001 (Unpaired two-tailed Student's t-test). B. CHOP mRNA levels (normalized to YWHAZ) in the thyroid glands of WT (+/+) and TG rdw/rdw rats (n=7-8 animals/group; each point =1 animal; mean ± S.D.) *** p < 0.001 (Unpaired two-tailed Student's t-test). C. Upper: Representative samples showing spliced and unspliced XBP1 mRNA in the thyroids of WT, TG rdw/+ and TG rdw/rdw rats (n=3-6 animals/group; each lane =1 animal). Hprt1 is a loading control. Lower: Quantitation of the fraction of spliced XBP1; mean ± S.D.) ** p < 0.01, *** p < 0.001 (One-way ANOVA, Bonferroni post-hoc test). D. Representative TUNEL staining and immunofluorescence of T 4 -containing protein with DAPI counter-stain in the thyroids of WT (+/+) and TG rdw/rdw rat (n=4 animals/group); scale bars = 20µm. E. Representative immunofluorescence of cleaved caspase-3 with DAPI counter-stain in thyroids of WT (+/+) and TG rdw/rdw rats (n=5 animals/group). For clarity, a dashed white line delimits the thyroid follicle lumen in the WT (in which cleaved caspase-3 is not detectable); scale bars = 20µm. F. Western blotting of PARP in thyroid glands from WT (+/+) and TG rdw/rdw rats (n=3-4; each lane =1 animal). G. Left panel: representative Western blotting of T 4containing protein in thyroid homogenates of WT and TG rdw/rdw rats (n=5 animals/group) ± soluble competitor T 4 to block specific bands (left of dotted red line). Right panel: the same samples immunoblotted with mAb anti-Tg showing intentional overloading of the TG rdw/rdw rat sample. Transmission EM survey of TG rdw/rdw rat thyroid follicles (scale bars = 2 µm). Panel C highlights engorged ER vacuoles in the basal cytoplasm with apically-displaced nuclei. Panel D-H highlight dead-cell ghosts in various thyroid follicles, each at a different stage of cellular disintegration within the follicle lumen. Panel G also highlights living thyrocytes with abundant apical microvilli, which have internalized material from the follicle lumen into endo-lysosomes. Panel H highlights that until new dead cells enter the follicle lumen, there is progressive clearance of cellular debris from the luminal cavity. I. Thyroid gland size (normalized to body weight) in a cohort of young versus older animals (open symbols = 8.9 ± 1.7 wk; closed symbols = 33.4 ± 2.6 wk; males shown as squares and females as circles). Data are shown as mean ± S.D.; *** p < 0.001 (two-way ANOVA, Bonferroni post hoc test). J. Total T 4 level in serum of WT (+/+) and TG rdw/rdw rats as a function of age (males shown as squares and females as circles). Data are mean ± S.D.; **p < 0.01, ***p < 0.001 (two-way ANOVA, Bonferroni post hoc test).