A Single Amino Acid Substitution Prevents Recognition of a Dominant Human Aquaporin-4 Determinant in the Context of HLA-DRB1*03:01 by a Murine TCR

Background Aquaporin 4 (AQP4) is considered a putative autoantigen in patients with Neuromyelitis optica (NMO), an autoinflammatory disorder of the central nervous system (CNS). HLA haplotype analyses of patients with NMO suggest a positive association with HLA-DRB1* 03:01. We previously showed that the human (h) AQP4 peptide 281–300 is the dominant immunogenic determinant of hAQP4 in the context of HLA-DRB1*03:01. This immunogenic peptide stimulates a strong Th1 and Th17 immune response. AQP4281-300-specific encephalitogenic CD4+ T cells should initiate CNS inflammation that results in a clinical phenotype in HLA-DRB1*03:01 transgenic mice. Methods Controlled study with humanized experimental animals. HLA-DRB1*03:01 transgenic mice were immunized with hAQP4281-300, or whole-length hAQP4 protein emulsified in complete Freund’s adjuvant. Humoral immune responses to both antigens were assessed longitudinally. In vivo T cell frequencies were assessed by tetramer staining. Mice were followed clinically, and the anterior visual pathway was tested by pupillometry. CNS tissue was examined histologically post-mortem. Flow cytometry was utilized for MHC binding assays and to immunophenotype T cells, and T cell frequencies were determined by ELISpot assay. Results Immunization with hAQP4281-300 resulted in an in vivo expansion of antigen-specific CD4+ T cells, and an immunoglobulin isotype switch. HLA-DRB1*03:01 TG mice actively immunized with hAQP4281-300, or with whole-length hAQP4 protein were resistant to developing a neurological disease that resembles NMO. Experimental mice show no histological evidence of CNS inflammation, nor change in pupillary responses. Subsequent analysis reveals that a single amino acid substitution from aspartic acid in hAQP4 to glutamic acid in murine (m)AQP4 at position 290 prevents the recognition of hAQP4281-300 by the murine T cell receptor (TCR). Conclusion Induction of a CNS inflammatory autoimmune disorder by active immunization of HLA-DRB1*03:01 TG mice with human hAQP4281-300 will be complex due to a single amino acid substitution. The pathogenic role of T cells in this disorder remains critical despite these observations.


Introduction
Neuromyelitis optica (NMO) is a demyelinating inflammatory disorder of the central nervous system (CNS) that is clinically and pathologically defined as the co-occurrence of optic neuritis and myelitis [1,2]. Aquaporin (AQP)4 is considered a potential autoantigen in patients with NMO after an autoantibody, designated NMO-IgG, that binds to human (h) AQP4 was detected in the serum of the vast majority of patients with NMO [3,4]. The presence of the NMO-IgG has led many neurologist and neuroimmunologists to believe that NMO may be a primarily B cell-mediated disease.
However, there is evidence to suggest a cellular immune response in NMO during disease initiation or perpetuation [5,6]. HLA haplotype analyses of patients with NMO suggest a positive association with HLA-DRB1 Ã 03:01 (DR17) [7,8] [9], a gene that codes for a major histocompatibility class (MHC) II molecule that presents linear antigens to CD4 + T cells [10]. Also, NMO-IgG is undetectable in a substantial number of patients with NMO [3]. A NMO-IgG, antibody isotype switch from IgM to IgG could not occur without CD4 + T cell involvement [11,12], which are abundantly present in NMO lesions [13]. B cell-depleting therapies are not consistently beneficial in patients with NMO [14][15][16]. Finally, transfer of AQP4-reactive T cells into wild-type mice and rats results in neurological deficits and CNS inflammation [17,18].
Other investigators have identified immunogenic linear determinants in various rodent species [5,6,[19][20][21]. We have previously shown that human AQP4 peptide 281-330 (hAQP4 281-300 ) is the dominant immunogenic determinant of hAQP4 in the context of HLA-DRB1 Ã 03:01 [21]. Characterizing the encephalitogenic role of these AQP4 specific T helper cells will bring to light the role of the cellular immune response in the initiation and progression of the NMO clinical disease phenotype.
In this study we intended to establish a T cell-mediated animal model of NMO in the context of HLA-DRB1 Ã 03:01, utilizing hAQP4 281-300 as the dominant hAQP4 determinant in that MHC II haplotype. Induction of an autoimmune disorder resembling experimental autoimmune encephalomyelitis (EAE) [22] was attempted by active immunization and adoptive transfer. Clinical disease activity, CNS tissue inflammation, and changes in pupillary reflexes were assessed. Alanine scanning of AQP4 281-300 was performed to test recognition by mouse T cell receptors (TCRs) and HLA-DRB1 Ã 03:01.
We were unable to induce clinical EAE, CNS inflammation, or altered pupillary responses. Disease resistance is the result of a single amino acid substitution from aspartic acid in hAQP4 to glutamic acid in murine (m)AQP4 at position 290 prevents the recognition of hAQP4 281-300 by the murine T cell receptor (TCR).

Results
Immunization with human (h)AQP4 281-300 leads to an expansion of antigen-specific CD4 + T cells in vivo Following immunization with human (h)AQP4 281-300 an expansion of antigen-specific CD4 + T helper cells was detected by tetramer staining of lymph node cells ( Fig 1A).
Immunization with human (h)AQP4 281-300 leads to an Ig isotype switch in HLA-DRB1*03:01 transgenic mice CD4 + T helper cells provide soluble mediators that drive B cell differentiation immunoglobulin (Ig) class switching. To determine whether hAQP4 281-300 -reactive CD4 + T cells are capable of causing IgM to IgG isotype switching in HLA-DRB1 Ã 03:01 transgenic mice, the concentration of Ig against hAQP4 281-300 , mAQP4284-299, or with whole-length hAQP4 protein in serum of immunized mice was quantified longitudinally. Since the NMO-IgG is a human IgG1 isotype, both, the murine IgG2a and IgG2b isotype were examined as they have similar properties with regard to complement binding and the Fcγ receptor. A switch from IgM to IgG2b was detected in mice immunized with hAQP4 281-300 peptide with regard to antibody responses against hAQP4 281-300 (Fig 1B), and whole-length AQP4 protein ( Fig 1C). An Ig isotype switch from IgM to IgG2b was also detectable in mice immunized with whole-length AQP4 protein with regard to antibody responses against hAQP4 281-300 (Fig 1D), and whole-length AQP4 protein ( Fig 1E). Thus, B cells of HLA-DRB1 Ã 03:01 transgenic mice are capable of recognizing hAQP4 281-300 peptide via the B cell receptor (BCR), and the cellular immune response against hAQP4 281-300 subsequently drives Ig isotype switching. These data support our previously published data that hAQP4 281-300 is a dominant determinant in HLA-DRB1 Ã 03:01 [21].

Active immunization with hAQP4 does not lead to clinical disease
We first examined whether active immunization of HLA-DRB1 Ã 03:01 transgenic [23] mice with hAQP4 results in clinical disease. A multitude of experimental procedures and conditions were tested to examine the encephalitogenic potential of hAQP4 peptides using the transgenic mice. Previously, our laboratory determined that hAQP4 281-300 was capable of generating a strong Th 1 and Th 17 immune response as measured by IFNγ and IL-17 ELISpot assay [21]. We performed active immunization with whole-length hAQP4 protein, hAQP4 281-300 , or mAQP4 281-300 in an attempt to generate an animal model of NMO. Experimental animals also received intraperitoneal (i.p.) injections of pertussis toxin (Ptx) on the day of immunization and two days post immunization [24]. This approach resulted in no observable clinical paralysis commonly seen in EAE models (Fig 2A). Immunization with a positive control proteolipid protein (PLP) 91-110 , the dominant encephalitogenic determinant in HLA-DRB1 Ã 03:01 [25] led to typical EAE (Fig 2A). All EAE experiments were terminated at day 30. None of the experimental animals immunized with PLP 91-110 that developed EAE died prematurely.
Subsequently, alternative methods to generate a T cell-mediated NMO model were employed. Some CNS autoimmune disease animal models require weekly booster immunization to generate disease phenotypes [26]. The rationale for this approach is to increase the encephalitogenic potential of hAQP4 281-300 by overcoming mechanisms of peripheral tolerance. In other experiments, additional booster immunization were given at day fourteen with adjuvants other than CFA, including QuilA and incomplete Freund's adjuvant (IFA) . Again, no clinical disease was observed (Fig 2A).

Adoptive transfer of hAQP4281-300-specific CD4 + T cells does not lead to clinical disease
In the adoptive transfer EAE model myelin-reactive activated CD4 + T cells are transferred into a naïve recipient. This model has some propensities that are very different from activelyinduced EAE: [27] The potential effects of adjuvant and pertussis toxin on the innate immune system are eliminated, and [27] in vitro re-activated donor T cells are less dependent on reactivation within the recipient CNS [28]. This model was specifically developed to test the role of antigen-specific donor T cells in EAE pathogenesis [29]. Since Th 1 and Th 17 cells have been shown in the EAE model to be capable of causing disease when passively transferred, we next examined whether hAQP4 281-300 -specific T cells could cause disease via adoptive transfer. No disease phenotype was detected ( Fig 2B). These results indicate that reactivation of hAQP4 281-300 -specific or mAQP4 281-300 -specific CD4 + Th 1 cells does not occur in the CNS of HLA-DRB1 Ã 03:01 recipient mice.
On histopathological examination there were no visible signs of cellular infiltration, inflammation, or demyelination within the brain and spinal cord in any experimental paradigms other than in active immunization with PLP 91-110 , the dominant encephalitogenic determinant in HLA-DRB1 Ã 03:01 that led to typical EAE (spinal cord shown in Fig 2C; inflammatory infiltrates and areas of demyelination are indicated by black arrows). The absence of any functional deficits was further corroborated through measuring of the pupillary reflex by murine pupillometry on day 15 post immunization ( Fig 2D). Mice did not show altered pupillary responses, further substantiating our previous findings that no functional or structural damage had occurred within the optic nerve.
The outcomes of these experiments suggested that mAQP4 281-300 cannot be recognized in the context of HLA-DRB1 Ã 03:01, or that hAQP 281-300 cannot be recognized by B.10 TCR.

Single Amino Acid difference leads to blocking of hAQP4-mediated T cell proliferation and differentiation
Within the immunogenic hAQP4 281-300 , there is a single amino acid mutation at position 290 between the human peptide and the mouse analog: An aspartic acid (D) in the human peptide to glutamic acid (E) in the mouse ( Fig 3A). Both are negatively charged acidic amino acids that Immunization with human (h)AQP4 281-300 leads to an expansion of antigen-specific CD4 + T cells in vivo, and an Ig isotype switch in HLA-DRB1*03:01 transgenic mice. (A) Following immunization with human (h)AQP4 281-300 , an expansion of antigen-specific CD4 + T helper cells was detected by tetramer staining of lymph node cells. The fluorescent signal of HLA-DRB1*03:01-loaded tetramers minus the fluorescent signal of empty HLA-DRB1*03:01 tetramers is shown. CD4 + T helper cells provide soluble mediators that drive B cell differentiation immunoglobulin (Ig) class switching. To determine whether hAQP4 281-300 -reactive CD4 + T cells are capable of causing IgM to IgG isotype switching in HLA-DRB1*03:01 transgenic mice, the concentration of Ig against hAQP4 281-300 , mAQP4284-299, or with whole-length hAQP4 protein in serum of immunized mice was quantified longitudinally. Since the NMO-IgG is a human IgG1 isotype, both, the murine IgG2a and IgG2b isotype were examined as they have similar properties with regard to complement binding and the Fcγ receptor. A switch from IgM to IgG2b was detected in mice immunized with hAQP4 281-300 peptide with regard to (B) antibody responses against hAQP4 281-300 and (C) whole-length AQP4 protein. An Ig isotype switch from IgM to IgG2b was also detectable in mice immunized with whole-length AQP4 protein with regard to (D) antibody responses against hAQP4 281-300 and (E) whole-length AQP4 protein.
doi:10.1371/journal.pone.0152720.g001 contain a carboxylic acid at the end of their side-chains. The difference between the two amino acids is an additional methyl group in the side chain of glutamic acid.
In lymph node cells of HLA-DRB1 Ã 03:01 mice immunized with hAQP4 281-300 there was a significant proliferation of both CD4 + T cells when 25 μg/ml of hAQP4 281-300 was used as the recall antigen ( Fig 3B). A dose of 5 μg/ml hAQP4 281-300 , or mAQP4 281-300 did not result in a significant proliferative response (Fig 3B). Lymph node cells of HLA-DRB1 Ã 03:01 mice immunized with mAQP4 281-300 did also not proliferate in response to mAQP4 281-300 , or hAQP4 281-300 at either dose ( Fig 3C) . An ELISpot assay revealed a significantly increased frequency of IFNγ and IL-17 producing lymph nodes cells from HLA-DRB1 Ã 03:01 mice immunized with there was a significant proliferation of CD4 + T cells when hAQP4 281-300 was used as the recall antigen (* = P-value = 0.01). Only a higher recall antigen dose of 25 μg/ml resulted in a significant increase in proliferation, whereas as a dose of 5 μg/ml did not. (C) There was no proliferative response to mAQP4 281-300 at either dose . (D) There is a significantly increased frequency of IFNγ and IL-17 producing lymph nodes cells from HLA-DRB1*03:01 mice immunized with hAQP4 281-300 by ELISpot assay when hAQP4 281-300 , and hAQP4 281-299 are used as recall antigens. However, we were unable to detect antigen specific IFNγ and IL-17 producing lymph nodes cells when mAQP4 281-300 , or the negative control hAQP4 66-79 were used as recall antigens (** = P-value < 0.01). (E) IFNγ and IL-17 producing lymph nodes cells from HLA-DRB1*03:01 mice immunized with mAQP4 281-300 were undetectable with any of the recall antigens.
Our observations suggest that the aspartic acid residue plays a critical role in either the presentation of hAQP4 281-300 in the context of HLA-DRB1 Ã 03:01, or in the recognition of hAQP4 281-300 by the B.10 TCR.
hAQP4 281-300 and mAQP4 281-300 binds to the HLA-DRB1*03:01 MHC II molecule With no detectable cellular immune response against mAQP4 281-300 in HLA-DRB1 Ã 03:01, we next examined whether the single amino acid mutation affected the anchoring of the mouse peptide to the HLA-DRB1 Ã 03:01 molecule. To identify critical residues of the AQP4 peptides, alanine-scanning peptides were generated that replaced each amino acid of mAQP4 281-300 with an alanine to aid in distinguishing anchor residues from contact residues ( Table 1). Since we previously identified hAQP4 284-299 to be the immunogenic region within hAQP4 281-299 [21], the alanine scanning peptides assessed only these residues.
First, the ability of hAQP4 281-300 -reactive lymph node cells to recognize the alanine screening peptides was determined by ELISpot (Fig 4A). Alanine screening peptides that not result in an increased frequency of IFNγ and IL-17 secreting lymph node cells were identified as the key residue peptides. Utilizing a flow-cytometry based MHC II binding assay, we were able to delineate between anchor residues and TCR contact residues. To perform the flow-cytometry based MHC II binding assay, peptides were biotinylated so that when presented in the context of HLA-DRB1 Ã 03:01, a FITC-avidin would distinguish peptides that were capable of being presented from those that could not. In comparing the percent of FITC-avidin positive cells, amino acids 288E and 294L were identified as the main anchor residues that interact with the HLA-DRB1 Ã 03:01 MHC II molecule (Fig 4B). The remaining residues, including the 290D residue that distinguishes mAQP4 281-300 and hAQP4 281-300 , were not required for binding. As a negative control, alanine scanning peptides were tested in C57BL/6 mice to examine the critical residues for binding to the H-2b MHC II molecule. Despite hAQP4 281-300 being able to be presented on the H-2b MHC II molecule, the critical residues were not similar to the critical residues necessary for binding to the HLA-DRB1 Ã 03:01 MHC II molecule. This observation may explain why we were unable to elicit cellular immune response against hAQP4 281-300 in C67BL/6 mice immunized with this peptide (data not shown). These data indicate that the mAQP4 281-300 binds to the HLA-DRB1 Ã 03:01 molecule via the same anchor residues and is thus able to be presented on the MHC II molecule in HLA-DRB1 Ã 03:01 transgenic mice.
AQP4 residue 290 mediates recognition by the B.10 TCR With mAQP4 281-300 being capable of being presented on the HLA-DRB1 Ã 03:01 molecule, we next examined whether the single amino acid mutation inhibited the contact with the hAQP4 281-300 specific B.10 TCR. Utilizing the alanine-scanning peptides, it was possible to identify critical residues that are required for generating the cellular immune response against the peptide. In performing IFNγ and IL-17 ELISpot assays with lymph node cells restimulated with alanine screening peptides (Table 1), E288A, T289A, D290A, D291A, I293A, and L294A were identified as critical amino acids for either the binding of the peptide to HLA-DRB1 Ã 03:01, or interacting the AQP4 281-300 specific T cell receptor (TCR). None of the other amino acids are required for either binding to HLA-DRB1 Ã 03:01 molecule, or recognition by the B.10 TCR.
Subsequently, utilizing a MHC binding assay, only the previously identified peptides E288A, T289A, D290A, D291A, I293A, and L294A were screened for their ability to bind to HLA-DRB1 Ã 03:01. 288E and 294L were identified to be HLA-DRB1 Ã 03:01 anchor residues, and the remaining residues play a critical role in the contact between the peptide and B.10 TCR (Fig 4B). These observations suggest that the D to E mutation between hAQP4 and mAQP4 peptides prevent the activation of hAQP4-specific T cells against mAQP4 281-300 .
We did not observe Ig class switching in mice immunized with mAQP4 284-299 against whole-length AQP4 protein (Fig 4C), substantiating our observation that immunization with mAQO4 284-299 does not drive antigen-driven T cell activation and a T cell proliferative response.

Discussion
In this investigation, we show that immunization with the immunogenic, hAQP4 281-300 determinant in HLA-DRB1 Ã 03:01 Tg mice, while leading to an expansion of antigen-specific CD4 + T cells, an Ig isotype switch of anti-hAQP4 281-300 antibodies, and a robust Th 1 and Th 17 immune response, does not lead to a clinical disease phenotype via active immunization or passive transfer of hAQP4 281-300 -specific CD4 + T cells. We also were unable to detect any evidence of electrophysiological anterior pathway pathology, or any evidence of CNS infiltration in our mouse model. It is conceivable that there may have been inflammatory infiltrates within the meninges of the brain and spinal cord. This was not assessed.
In summary, our data suggest that a single amino acid substitution between hAQP4 and mAQP4 is one possible reason why it will be challenging to establish an animal model of NMO in HLA-DRB1 Ã 03:01 transgenic mice. For induction of CNS autoimmunity, recognition of the cognate antigen by the host immune system is an absolute requirement [24,28]. Within the pathogenic AQP4 281-300 , the glutamic acid (D) to aspartic acid (E) mutation results in the addition of a methyl group within the side chain of the AQP4 287 residue in the murine peptide. Despite being the same polarity, the hAQP4 281-300 -specific TCR can differentiate between the human and the mouse peptides. This was corroborated with data showing that the AQP4 287 residue was important for contact with the B.10 TCR rather than being a MHC II anchor residue. Furthermore, antigen recall with mAQP4 281-300 in HLA-DRB1 Ã 03:01 mice immunized with mAQP4 281-300 does not result in proliferation of CD4 + T cells. Likely, negative thymic selection for mAQP4 281-300 specific CD4 + T cells occurs in these mice, and prevents active disease induction.
T cell mediated disease models of NMO will be necessary to fully understand the complexity of this disorder. Induction of a CNS inflammatory autoimmune disorder by active immunization of HLA-DRB1 Ã 03:01 TG mice with human hAQP4 281-300 will be complex due to a single amino acid substitution.
It is also possible that hAQP4 281-300 , the dominant determinant of hAQP4 in HLA-DRB1 Ã 03:01 in vitro, does not result in generation of encephalitogenic CD4 + T cells in vivo. This possibility has not conclusively been ruled-out.
The pathogenic role of T cells in this disorder remains critical despite these observations.

Animal model
For all experiments, individual animals were observed daily based on the EAE clinical scoring system as follows: 0 = no clinical disease, 1 = loss of tail tone, 2 = mild paraparesis, 3 = paraplegia, 4 = hindlimb and forelimb paralysis, 5 = moribund or death. Observation of all experimental animals occurred at least twice daily, with documentation of the clinical score. The following interventions are cumulative: Once a mouse reached clinical score 2, moist chow was provided daily. At score 3, animal weights were recorded daily, and animals were euthanized if the weight loss was greater than 20% from baseline. When mice scored 4, the urinary bladder was palpate and manually expressed as needed, and affected animals were no longer housed with cagemates of a lower score. Furthermore, suitable nesting material was always be provided. Affected mice had to be euthanized if there is no improvement after 72 hours at score of 4. Animals were euthanized immediately upon observation of a score of 5, regardless of time to development. Euthanasia was performed by carbon dioxide narcosis. Cervical dislocation was always used as a secondary physical method. Death was confirmed by observing for lack of breathing, loss of heartbeat, glazed appearance to the eyes, and loss of limb movement.

Histology
Following fixation in 10% buffered formalin, coronal sections of brain tissue, axial sections of spinal cord, and longitudinally-oriented optic nerves were processed and embedded in paraffin blocks. 4 μm sections were cut, mounted on Fisher Brand Superfrost Plus glass slides (Fisher Scientific, Pittsburgh, PA), and stained with hematoxylin & eosin (Fisher Scientific).
For the Luxol Fast Blue stained-sections, 6 μm thick sections of paraffin-embedded tissue were cut on a rotary microtome and mounted on Fisher Brand Superfrost Plus glass slides (Fisher Scientific). The sections were deparaffinized and hydrated. Following heating of the sections in 0.1% Luxol Fast Blue (Sigma-Aldrich, St. Louis, MO) at 60°C for at least one hour, excess stain was rinsed off. They were then differentiated in 70% alcohol for 25 seconds and rinsed in distilled water. Next, the sections were evaluated under the microscopy and depending on the adequacy of differentiation they were subjected to an additional round of lithium carbonate, and an alcohol rinse.

Pupillometry
The pupillary reflex of experimental mice were measured using the pupillometry system by Neuroptics Inc. (San Clemente, CA) previously described in Husain et al [32]. Briefly, infrared cameras capture digital images of mouse pupils in darkness at a baseline level after sedation. After the baseline pupil size is determined, an intensity-calibrated light source emits a stimulus into one or both of the eyes and a custom program measures the pupil diameter designed to analyze pupil size, onset latency, constriction velocity, and response amplitude. The light stimulus consists of a flash of light at 2, 32, or 125 μWs.

Tetramer analysis
The frequency of antigen specific CD4 + T-cells was assessed using a tetramer binding assay protocol provided by the Benaroya Research institute [33]. Briefly, lymph nodes were harvested, and a single cell suspension was generated. Cells were resuspended in media, and HLADRB1 Ã 03:01 tetramers loaded with hAQP4 284-298 , or empty HLADRB1 Ã 03:01 tetramer (Benaroya Research Institute, Seattle, WA) were added, and incubated for 90 minutes at 37°C and 5%CO 2 . Cells were then washed with FACS buffer and blocked with anti-CD16/CD32, and stained following the protocol for cell surface staining described above with anti-CD3 APC (17A2, Tonbo Biosciences, San Diego, CA) and anti-CD4-Pacific Blue (RM4-5, BD-Biosciences). Data was acquired with a FACS Canto RUO II Special Order (BD Biosciences), and data was analyzed using FlowJo software (Tree Star).

MHC binding assay
This assay was adapted from a protocol found in Busch et al [34]. Briefly, spleens isolated from naïve HLADRB1 Ã 03:01 transgenic mice were used to generate single cell suspensions. Then, 1x10 6 splenocytes were incubated with either biotinylated hAQP4 281-300 , mAQP4 281-300 , or hAQP4 284-299 alanine screening peptides at a concentration of 10μ/mL for 4 hours at 37°C in a 96-well plate. Post incubation, cells were washed two times with FACS buffer and stained for flow cytometry utilizing avdin-FITC (Biolegend) applying the previously described protocol. After staining cells with avidin-FITC, they were run through a Accuri C6 (BD) flow cytometer, or a BD FACSCalibur, and Flowjo was utilized to quantify the percentage of FITC positive cells. A potential delta between the control peptides (hAQP4 284-299 , and mAQP4 281-300 ), and the alanine screening peptides is considered to be the result of the alanine substitutions.

Statistical Analysis
For parametric tests, data were checked for normality by using the Kolmogorov-Smirnov test.
Normally distributed values were compared using the unpaired two-sided Student t-test. Correlations between continuous and categorical variables were assessed using the Mann-Whitney U-test. All experiments were repeated at least twice. All statistical tests were 2-sided and p < 0.05 indicated significance. All analyses were performed with Prism 5 (Graphpad, La Jolla, CA, USA).