Sex differences in Alzheimer's disease plasma biomarker levels and clinical utility

Abstract

INTRODUCTION

Sex differences in Alzheimer's disease (AD) plasma biomarkers remain understudied despite higher AD risk in women.

METHODS

We examined sex differences in plasma amyloid beta (Aβ)42/40, phosphorylated tau (p‐tau)217, p‐tau217/Aβ42, glial fibrillary acidic protein (GFAP), and neurofilament light chain (NfL) in cognitively unimpaired (CU) and cognitively impaired (CI) Alzheimer's Disease Neuroimaging Initiative participants. For Aβ42/40, p‐tau217, and p‐tau217/Aβ42, we evaluated amyloid positron emission tomography positivity classification performance and associations with cognitive trajectories using sex interactions and sex‐stratified models.

RESULTS

Among CU participants, men had lower Aβ42 and GFAP, and higher p‐tau217/Aβ42. Among the CI group, GFAP, p‐tau217 and p‐tau217/Aβ42 were higher in women. Overall classification performance was similar across sexes; however, p‐tau217 and p‐tau217/Aβ42 showed higher specificity and positive predictive value in CU women, with the opposite pattern observed in CI participants. In CU participants, p‐tau217 and p‐tau217/Aβ42 predicted modified Preclinical Alzheimer Cognitive Composite decline only in women.

DISCUSSION

Sex‐specific plasma biomarker cutoffs may not be necessary. However, sex influences biomarker levels, classification metrics, and prognostic value, highlighting the importance of considering sex differences when interpreting biomarker results and optimizing trial enrichment strategies.

Highlights

• Cognitively unimpaired (CU) men have lower plasma amyloid beta (Aβ)42 while cognitively impaired (CI) women have higher plasma phosphorylated tau (p‐tau)217.

• Plasma glial fibrillary acidic protein is elevated in women in both CU and CI groups.

• Overall performance for classifying amyloid positron emission tomography positivity is similar across sexes.

• P‐tau217 and p‐tau217/Aβ42 specificity and positive predictive value vary by sex and cognitive stage

• In the CU group, a positive plasma biomarker status predicts cognitive decline only in women.

1. BACKGROUND

Implementation of reliable plasma biomarkers for Alzheimer's disease (AD) would streamline diagnostics, reduce cost, and improve patient management through precision medicine. ¹ From a research perspective, it is paramount to better understand how these plasma biomarkers can be optimally incorporated into clinical trial design. As the field shifts toward a biological definition of AD and emphasizes early intervention, ² plasma biomarkers are key for identifying individuals suitable for clinical trials and disease‐modifying treatment. This includes identifying cognitively unimpaired (CU) individuals with amyloid pathology for prevention trials and selecting symptomatic patients with a specific disease burden who are most likely to respond to disease‐modifying treatments.

Key plasma biomarkers, particularly amyloid beta (Aβ)42/40 and phosphorylated tau 217 (p‐tau217), accurately detect the amyloid pathology characteristic of AD. ³ , ⁴ , ⁵ , ⁶ The p‐tau217/Aβ42 ratio also closely correlates with AD pathology and predicts disease progression. ⁷ , ⁸ , ⁹ Assays combining these biomarkers have recently received US Food and Drug Administration clearance or are under regulatory review for detecting brain amyloid pathology. ¹⁰ , ¹¹ , ¹² Other important biomarkers include glial fibrillary acidic protein (GFAP), which is closely associated with amyloid pathology, ¹³ , ¹⁴ , ¹⁵ and neurofilament light chain (NfL), an established marker of non‐specific neurodegeneration. ¹⁶ , ¹⁷ Despite their promise, the reported performance of these biomarkers varies considerably among studies, potentially influenced by clinical stage and individual traits.

Sex is an important factor in AD, with well‐recognized differences spanning pathophysiology and clinical presentation. ¹⁸ , ¹⁹ Women show greater vulnerability to certain AD risk factors, including apolipoprotein E (APOE) ε4, ²⁰ , ²¹ , ²² and exhibit higher burdens of tau pathology, ²² , ²³ , ²⁴ , ²⁵ , ²⁶ neurodegeneration, ²⁷ , ²⁸ and faster disease progression. ²⁷ , ²⁹ , ³⁰ , ³¹ However, women may have a greater resilience to the initial pathological changes of AD in its early stages. ³² , ³³ , ³⁴

Despite these known differences, research into how sex affects AD plasma biomarkers remains limited and has produced conflicting results. ³⁵ Most studies found no sex differences in plasma Aβ42/40, ³⁶ , ³⁷ although one study reported lower levels specifically in CU women. ³⁷ Meanwhile, some studies have reported no sex differences in plasma p‐tau species, ³⁶ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² NfL, ¹⁶ or GFAP, ¹⁴ , ⁴³ and others have found higher levels of p‐tau217 in CU men, ⁴⁰ or higher plasma GFAP ¹³ , ¹⁵ , ⁴⁴ in women. Notably, a recent study reported a sex‐specific association between plasma p‐tau217, verbal memory, and medial temporal lobe atrophy in CU women. ⁴² Research in autosomal dominant AD also suggested that sex‐specific differences in plasma p‐tau217 and NfL may affect cognitive performance. ⁴⁵

Although the source of discrepancies in plasma biomarker differences by sex is unclear, one possibility is that they are driven by sex differences in peripheral physiology that may influence biomarker concentrations independently of central nervous system pathology. Factors such as body mass index (BMI), renal function, plasma protein concentrations, immune function, or hormonal profiles, often differ between men and women ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ and can significantly influence analyte concentrations, possibly contributing to sex‐specific variations in plasma biomarker levels. ⁵⁰

Understanding how sex influences not only plasma biomarker levels but their classification and prognostic performance is critical for optimizing their use. Sex‐specific variations could impact the identification of individuals with underlying amyloid pathology or those at the greatest risk of progression—precisely those benefiting most from disease‐modifying therapies. Recent clinical trial results have suggested sex differences in treatment efficacy, ⁵¹ , ⁵² raising the possibility that this variability stems partially from how biomarkers identified and characterized men and women. Investigating these sex effects could therefore improve clinical trial enrichment, ensure equitable treatment evaluation, and enhance biomarker precision in research and practice.

This study aimed to evaluate potential sex differences in key AD plasma biomarkers, and their implications for detecting AD pathology and predicting disease progression. We investigated how sex impacts biomarker levels, their accuracy in identifying amyloid positron emission tomography (PET)‐positive individuals, and their association with cognitive decline. Focusing on the preclinical and prodromal disease stages, our analysis included CU and cognitively impaired (CI) participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study. Unlike most studies that treat sex as a biological confounding variable, we investigated sex as a direct modifier of plasma biomarker performance, analyzing their utility from a sex‐specific perspective.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature on sex differences in Alzheimer's disease (AD) plasma biomarkers, focusing on amyloid beta 42/40, phosphorylated tau 217, glial fibrillary acidic protein, and neurofilament light chain. While sex differences in AD pathology and clinical progression are well established, prior studies on plasma biomarkers have rarely examined whether classification performance or prognostic value differ between men and women.

2. Interpretation: Our results suggest that sex‐specific cutoffs may not be necessary for identifying amyloid positron emission tomography positivity using plasma biomarkers. However, sex differences were observed in plasma biomarker levels and in their association with cognitive trajectories, which should be considered when interpreting plasma biomarker results and their clinical implications.

3. Future directions: Future studies should investigate mechanisms underlying sex differences in plasma biomarker profiles, including hormonal, metabolic, and vascular contributors. Validation in diverse cohorts and exploration of sex‐informed analytical strategies may improve biomarker interpretation and help optimize participant selection and outcome monitoring in clinical trials.

2. METHODS

2.1. Study participants

ADNI participants classified as CU (Clinical Dementia Rating [CDR] = 0) or CI (CDR > 0) with available Fujirebio plasma Aβ42/40 and p‐tau217, and Quanterix NfL and GFAP measurements were included (N = 1508; N = 768 CU and N = 740 CI). A subset of 815 individuals (N = 466 CU and N = 349 CI) had available C2N Aβ42/40, and 462 individuals (N = 204 CU and N = 258 CI) had available C2N p‐tau217 and %p‐tau217.

For cross‐sectional analyses assessing the performance of plasma biomarkers to classify amyloid PET positivity, we included participants who had an amyloid PET scan within 1 year of plasma sample collection (N = 1119 [N = 593 CU and N = 526 CI] with Fujirebio plasma biomarkers, and N = 646 [N = 368 CU and N = 278 CI] for C2N plasma biomarkers). For longitudinal analyses evaluating the association of plasma biomarkers with cognitive decline, we selected the earliest available plasma sample to maximize the duration of follow‐up, including up to 5 years of annual cognitive assessments. This resulted in N = 1051 (N = 504 CU and N = 547 CI) individuals with Fujirebio plasma biomarkers and N = 492 (N = 222 CU and N = 270 CI) individuals with C2N plasma biomarkers. In the Fujirebio sample, participants had a median of three observations (interquartile range [IQR] 2–5), with 66.2% having three or more observations. In the C2N sample, participants had a median of five observations (IQR 3–6), with 81.3% having three or more observations.

The diagnostic criteria for ADNI participants have been described previously. ⁵³ Ethno‐racial background and sex were self‐reported, and APOE genotyping was performed as part of the standard ADNI protocol.

2.2. Plasma biomarker measurements

Plasma samples were collected, processed, and stored according to the standardized ADNI protocols. ⁵⁴ Analyses were performed using the following assays: Fujirebio Diagnostics Lumipulse (Fujirebio; Aβ42, Aβ40, p‐tau217), C2N Diagnostics PrecivityAD2 (C2N; Aβ42, Aβ40, p‐tau217, and %p‐tau217), and Quanterix Neurology 4‐Plex (Quanterix) assays (NfL and GFAP). ⁵ A total of 376 (43.7%) C2N PrecivityAD2 p‐tau217 measurements fell below the assay's reliable detection range and were imputed. To minimize bias, these imputed values were excluded except for binary classification of participants (positive/negative), for which the imputed p‐tau217 values did not influence the classification outcome. Corresponding %p‐tau217 values derived from imputed p‐tau217 measurements were excluded from all analyses.

2.3. Amyloid and tau PET acquisition, processing, and status determination

Amyloid PET imaging with ¹⁸F‐florbetapir or ¹⁸F‐florbetaben was conducted at each ADNI site adhering to standardized protocols. ⁵⁵ Global standardized uptake value ratios (SUVRs) were calculated by averaging across cortical regions (frontal, cingulate, parietal, and lateral temporal) defined by FreeSurfer version 7.1. SUVRs were normalized to whole cerebellum and Centiloid values were derived using the transformations provided by the ADNI PET Core. ⁵⁶ A Centiloid value of 25 was used to define amyloid PET positivity.

Tau PET imaging with ¹⁸F‐flortaucipir was conducted at each ADNI site following standardized protocols. SUVRs were calculated for a mesial‐temporal (entorhinal, parahippocampus, and amygdala) meta‐region of interest (ROI) defined by FreeSurfer version 7.1. SUVRs were normalized to the inferior cerebellar gray matter reference region, using the SUIT template. ⁵⁷

2.4. Cognitive assessments

Participants were classified as CU if they had a CDR global score of 0 and as CI if they had a CDR global score > 0. To assess the association of plasma biomarkers with cognitive trajectories, we used cognitive outcomes commonly used in AD clinical trials: the Digit Symbol Substitution Test version of the modified Preclinical Alzheimer Cognitive Composite (mPACC) for CU individuals, ⁵⁸ and the CDR Sum of Boxes (CDR‐SB) ⁵⁹ and Alzheimer's Disease Assessment Scale Cognitive subscale consisting of 13 items (ADAS‐Cog13) ⁶⁰ for CI individuals.

2.5. Statistical analyses

All analyses were run separately for CU and CI participants. Sex differences in age and BMI were assessed using a t test, and Pearson chi‐squared (χ ²) was used to compare frequency of APOE ε4 carriership (carrier vs. non‐carrier), education attainment, ethno‐racial background, cognitive status, and frequency of comorbidities. Differences in cognitive scores were tested with one‐way analysis of covariance (ANCOVA) adjusted by age, the presence of at least one APOE ε4 allele (APOE ε4 status), and education attainment. Sex differences in plasma biomarker levels were evaluated using three different models: (1) unadjusted one‐way analysis of variance (ANOVA); (2) ANCOVA adjusted by age; and (3) ANCOVA adjusted by age and BMI, given previous evidence of BMI influence on plasma biomarker levels. ⁴⁸ , ⁴⁹ Additionally, we assessed sex differences in common comorbid conditions (dyslipidemia, diabetes, hypertension, chronic kidney disease, and obesity), and models were further adjusted for these conditions when their prevalence differed by sex. Finally, age‐adjusted ANCOVA was used to examine the potential interactions between sex and APOE ε4 status, or sex and amyloid PET status on plasma biomarker levels.

We performed receiver operating characteristic (ROC) curve analysis to evaluate the accuracy of plasma biomarkers for identifying amyloid PET positivity. The analysis was conducted on the entire sample and then stratified by sex. The resulting areas under the curve (AUCs) were compared using the DeLong test. For our primary analysis, we determined an optimal amyloid PET positivity classification threshold for the entire sample by selecting the cutoff that maximized the sum of sensitivity and specificity, also known as the Youden index. At this threshold, we calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), with 95% confidence intervals derived from 1000 bootstrap resamples. To assess performance consistency, this common threshold was applied to men and women separately. We then derived sex‐specific optimal thresholds to determine whether they improved classification performance compared to the common threshold.

We also conducted a two‐cutoff analysis to improve classification confidence. We established a high‐sensitivity threshold (maximum specificity with ≥ 90% sensitivity) and a high‐specificity threshold (maximum sensitivity with ≥ 90% specificity). Cases with values between these two cutoffs were classified as “intermediate.” ⁶¹ , ⁶²

For all analyses, sex differences in accuracy, sensitivity, and specificity were examined using the Pearson χ ² test, while differences in PPV and NPV were evaluated using permutation tests (10,000 iterations). Additionally, sex differences in the proportion of “intermediate” cases were compared using a Pearson χ ² test.

Finally, to examine sex differences in the association between baseline plasma biomarker–based amyloid positivity and cognitive decline, we first classified participants as plasma biomarker positive or negative using the binary cutoffs derived from the whole sample. We then fit linear mixed effects models with random intercepts and slopes, including the three‐way interaction among sex, time, and biomarker group (positive/negative). In addition, separate models evaluated the main effect of biomarker group on cognitive decline (time by biomarker group interaction) in the overall sample and within sex‐stratified subsamples. Models were adjusted for the interactions of time with age, APOE ε4 status, and education attainment. For reference, these analyses were also performed using amyloid PET positivity stratification.

For plasma biomarkers that showed significant associations with cognitive decline in the primary analyses, we conducted post hoc sensitivity analyses. These included additional adjustment for plasma GFAP to assess whether these associations were independent of differences in astroglial reactivity, and separate models adjusting for BMI, given prior reports of paradoxical associations between higher BMI and cognitive outcomes, particularly in women. ⁶³ Finally, to ensure that the inclusion of participants with mild dementia (CDR ≥ 1) did not confound our findings, we performed a sensitivity analysis restricted to participants with mild cognitive impairment (MCI; CDR = 0.5).

Of note, analyses of classification performance and associations with cognitive decline focused on core AD plasma biomarkers (Aβ42/40, p‐tau217, and p‐tau217/Aβ42) due to their relevance for AD clinical trial enrichment; GFAP and NfL were excluded from these specific utility analyses.

For all longitudinal analyses, follow‐up cognitive data was restricted to up to 5 years, to ensure that cognitive trajectories could be reasonably modeled using linear methods. In regression analyses, normality of residuals was tested with visual inspection of the histograms. All tests were two tailed, with a significance level of α  =  0.05. p values reported in the text and tables correspond to unadjusted values. To control for multiple comparisons, the false discovery rate (FDR) ⁶⁴ method was applied across the number of plasma biomarkers tested. In the text, only results that remained statistically significant after FDR correction are discussed, unless otherwise specified. In tables and figures, p values surviving FDR correction are denoted with asterisks. Statistical analyses and figures were performed in open‐source statistical software R v.4.1.2.

3. RESULTS

3.1. Participants’ baseline characteristics and sex differences in plasma biomarker levels

Participants’ characteristics by cognitive status and sex are shown in Table 1. Compared to women, men were older and had higher education attainment in both CU and CI groups, and the proportion of non‐Hispanic Black individuals was higher in women in both CU and CI groups. Within CU individuals, men presented with lower cognitive performance in their Mini‐Mental State Examination, mPACC, and ADAS‐Cog13 scores, while no differences in cognitive performance between men and women were found in the CI group. There were no sex differences in the frequency of common comorbidities aside from a higher prevalence of dyslipidemia in CU men compared to CU women. BMI was significantly higher in CI men compared to women.

TABLE 1.

Participant characteristics and biomarker levels by cognitive status and sex.

	Cognitively unimpaired					Cognitively impaired
	Men		Women			Men		Women
	n	Value	N	Value	p value	n	Value	n	Value	p value
Age, years	330	76.1 ± 7.75	438	72.3 ± 8.70	<0.0001	408	77.3 ± 8.05	332	75.1 ± 8.82	0.0003
APOE ε4 carriers, n (%)	330	106 (32.1)	438	149 (34.0)	0.63	408	204 (50.0)	332	179 (53.9)	0.32
Ethno‐racial background (non‐Hispanic White/non‐Hispanic Black/other), n (%)	330	268 (81.2) / 31 (9.40) / 31 (9.40	438	314 (71.7) / 87 (19.9) / 37 (8.40)	0.0004	408	375 (91.9) / 13 (3.20) / 20 (4.90)	332	261 (78.6) / 47 (14.2) / 24 (7.20)	<0.0001
Education attainment (high school or less/college/postgraduate), n (%)	330	19 (5.80) / 122 (37.0) / 189 (57.3)	438	45 (10.3) / 191 (43.6) / 202 (46.1)	0.004	408	63 (15.4) / 173 (42.4) / 172 (42.2)	332	53 (16.0) / 168 (50.6) / 111 (33.4)	0.042
CDR Global 0/0.5/ ≥ 1, n (%)	330	330 (100.0) / 0 (0.0) / 0 (0.0)	438	438 (100.0) / 0 (0.0) / 0 (0.0)	—	408	0 (0.0) / 259 (63.5) / 149 (36.5)	332	0 (0.0) / 242 (72.9) / 90 (27.1)	0.008
MMSE	328	28.7 ± 1.54	434	29.1 ± 1.14	0.0013	404	24.2 ± 5.45	331	24.6 ± 5.52	0.32
CDRSB	330	0.05 ± 0.16	438	0.05 ± 0.15	0.52	408	3.97 ± 3.46	332	3.40 ± 3.50	0.089
ADAS‐Cog13	328	11.0 ± 4.87	427	8.17 ± 4.61	<0.0001	390	24.1 ± 12.8	315	22.3 ± 13.4	0.14
mPACC	185	−0.89 ± 3.27	222	0.57 ± 2.94	<0.0001	301	−12.3 ± 10.9	209	−12.7 ± 9.31	0.74
BMI, kg/m2	326	27.3 ± 4.24	432	27.4 ± 6.13	0.72	401	27.0 ± 4.27	323	26.1 ± 5.49	0.019
Dyslipidemia, n (%)	135	59 (17.9)	112	41 (9.40)	0.0008	207	70 (17.2)	133	64 (19.3)	0.52
Diabetes, n (%)	135	8 (5.90)	112	7 (6.20)	0.99	207	22 (10.6)	133	9 (6.80)	0.31
Hypertension, n (%)	327	116 (35.5)	435	141 (32.4)	0.42	405	143 (35.3)	325	107 (32.9)	0.55
Chronic kidney disease, n (%)	135	37 (27.4)	112	28 (25.0)	0.78	207	53 (25.6)	133	35 (26.3)	0.98
Obesity, n (%)	326	73 (22.5)	432	122 (28.6)	0.067	401	73 (18.2)	323	69 (21.4)	0.33
Amyloid PET, Centiloid	250	18.7 ± 34. 7	343	22.2 ± 34.5	0.025	273	53.3 ± 53. 9	253	54.4 ± 50.7	0.49
Amyloid PET positive, n (%)	252	78 (31.0)	346	110 (31.8)	0.90	282	163 (57.8)	254	152 (59.8)	0.70
Mesial‐temporal tau PET SUVR	194	1.19 ± 0.19	270	1.21 ± 0.22	0.003	199	1.53 ± 0.47	189	1.69 ± 0.65	0.004
Plasma biomarkers
Fujirebio Aβ42, pg/mL	330	26.4 ± 8.86	438	27.1 ± 5.79	0.008*	399	26.5 ± 6.07	331	25.9 ± 7.26	0.83
C2N Aβ42, pg/mL	194	44.6 ± 8.66	272	46.3 ± 11.3	0.009*	168	43.5 ± 10.13	181	44.6 ± 9.84	0.15
Fujirebio Aβ40, pg/mL	330	308 ± 119	438	308 ± 66.5	0.10	399	321 ± 72.3	331	319 ± 94.3	0.34
C2N Aβ40, pg/mL	194	452 ± 91.0	272	457 ± 99.1	0.040	168	465 ± 98.8	181	464 ± 91.5	0.57
Fujirebio Aβ42/40	330	0.090 ± 0.013	438	0.091 ± 0.013	0.18	399	0.085 ± 0.011	331	0.085 ± 0.012	0.041
C2N Aβ42/40	194	0.095 ± 0.011	272	0.098 ± 0.015	0.30	168	0.094 ± 0.012	181	0.098 ± 0.014	0.047
Fujirebio p‐tau217, pg/mL	330	0.21 ± 0.25	438	0.17 ± 0.17	0.24	402	0.43 ± 0.36	331	0.52 ± 0.47	0.004*
C2N p‐tau217, pg/mL	86	2.77 ± 1.60	118	2.89 ± 1.68	0.41	125	4.45 ± 3.16	133	5.85 ± 4.21	0.003*
C2N %p‐tau217, %	86	4.46 ± 1.92	118	4.61 ± 2.45	0.59	125	7.15 ± 4.84	133	8.52 ± 5.13	0.036
Fujirebio p‐tau217/Aβ42	330	0.007 ± 0.009	438	0.006 ± 0.006	0.004*	402	0.016 ± 0.015	331	0.019 ± 0.022	0.001*
Quanterix NfL, pg/mL	330	20.9 ± 13.6	438	18.2 ± 10.4	0.85	402	28.0 ± 15.6	331	26.2 ± 15.6	0.88
Quanterix GFAP, pg/mL	330	155 ± 84.7	438	168 ± 95.6	<0.0001*	402	210 ± 113	331	245 ± 135	<0.0001*

Notes: Data are shown as mean ± standard deviation or n (%). t tests were used to evaluate sex differences in continuous variables and Pearson chi‐squared (χ ²) test for categorical variables. Amyloid PET Centiloid values, mesial‐temporal tau PET SUVR, or plasma biomarkers were compared with analyses of covariance adjusting by age. Models including cognition were adjusted by age and education attainment. Amyloid PET positivity was established at 25 Centiloids. Amyloid PET within 1 year of plasma collection was available for N = 1119 participants with Fujirebio plasma biomarkers (N = 593 CU and N = 526 CI) and for N = 646 individuals with C2N plasma biomarkers (N = 368 CU and N = 278 CI). BMI, obesity, and hypertension data correspond to measurements obtained within 1 year of plasma collection. Participants were classified as obese when BMI > 30 kg/m², and as hypertensive when systolic blood pressure ≥ 130 mmHg and diastolic blood pressure ≥ 80 mmHg. Medical history of chronic kidney disease, diabetes, and dyslipidemia was collected at study entry. For sex differences in plasma biomarker levels, asterisks (*) indicate statistical significance after FDR correction for multiple comparisons.

Abbreviations: Aβ amyloid beta; ADAS‐Cog13, Alzheimer's Disease Assessment Scale 13‐item Cognitive subscale; APOE, apolipoprotein E; BMI, body mass index; CDR, Clinical Dementia Rating; CDRSB, Clinical Dementia Rating Sum of Boxes; CI, cognitively impaired; CU, cognitively unimpaired; FDR, false discovery rate; GFAP, glial fibrillary acidic protein; MMSE, Mini‐Mental State Examination; mPACC, modified Preclinical Alzheimer Cognitive Composite; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau; SUVR, standardized uptake value ratio.

Regarding AD biomarker levels, after accounting for the effect of age, CU women had higher amyloid PET Centiloid values compared to men (β = 0.19, p = 0.025), although there were no statistically significant sex differences in the frequency of amyloid PET positivity (p = 0.90). Mesial‐temporal tau PET uptake was higher in women compared to men in both CU (β = 0.28, p = 0.003) and CI (β = 0.30, p = 0.004) groups.

For plasma biomarkers, we first evaluated the agreement between the two analytical platforms (Fujirebio and C2N) to ensure consistency. We observed strong correlations between assays measuring the same analyte, particularly for p‐tau217 (Spearman rho = 0.58 in CU individuals and 0.91 in CI individuals; Figures S1 and S2 in supporting information), confirming that both platforms capture similar pathological signals. GFAP was elevated in women compared to men in both CU (β = 0.36, p < 0.0001) and CI (β = 0.38, p < 0.0001) groups. CU men had a higher Fujirebio p‐tau217/Aβ42 ratio than women (β = −0.21, p = 0.004), and lower Aβ42 levels (β = 0.29, p = 0.008 for Fujirebio and β = 0.24, p = 0.009 for C2N Aβ42, respectively), while there were no differences in p‐tau217 measures (Figure 1; Table 1). Conversely, in the CI group, women had a higher Fujirebio p‐tau217/Aβ42 ratio (β = 0.22, p = 0.001) and higher levels of p‐tau217 measures (β = 0.22, p = 0.004 for Fujirebio p‐tau217 and β = 0.37, p = 0.003 for C2N p‐tau217, respectively). Sex differences in C2N %p‐tau217 and in C2N and Fujirebio Aβ42/40 levels were also observed in CI participants, although they did not survive multiple comparison correction (Figure 1, Table 1, and Table S1 in supporting information).

FIGURE 1.

Alzheimer's disease plasma biomarker levels by sex in cognitively unimpaired and cognitively impaired individuals. Box plots show median (horizontal bar), IQR (hinges), and 1.5 × IQR (whiskers). Asterisks indicate significant differences in plasma biomarker levels between men and women tested with an analysis of covariance adjusted for age, after false discovery rate multiple comparison correction. ** p < 0.01; ***p < 0.001. Aβ, amyloid beta; GFAP, glial fibrillary acidic protein; IQR, interquartile range; NfL, neurofilament light chain; p‐tau, phosphorylated tau

Sex differences in plasma Fujirebio and C2N Aβ42, Fujirebio p‐tau217/Aβ42 ratio, and GFAP in CU were maintained after further adjustment for BMI (Table S1). Results also remained unchanged when accounting for the higher frequency of dyslipidemia observed in CU men (β = 0.24, p = 0.011 for C2N Aβ42; β = 0.19, p = 0.008 for Fujirebio Aβ42; β = −0.20, p = 0.006 for Fujirebio p‐tau217/Aβ42 ratio; β = 0.36, p < 0.0001 for GFAP). In the CI group, the observed sex differences in plasma biomarkers were maintained after adjusting for BMI except for the nominal difference in C2N Aβ42/40 (p = 0.055; Table S1).

Notably, sex did not modify the association of any plasma biomarker with APOE ε4 status or amyloid PET status, suggesting the observed differences in plasma biomarker levels are independent of these factors (Table S2 in supporting information).

3.2. Sex differences in identifying amyloid PET–positive individuals

We evaluated whether the classification performance of core AD plasma biomarkers (Aβ42/40, p‐tau217, and p‐tau217/Aβ42) for identifying amyloid PET positivity differed by sex. First, a single optimal threshold for each biomarker was derived from the entire cohort. The performance of this common threshold was then assessed across the entire cohort, as well as separately within men and women, to determine whether the model performed similarly between sexes. We found no significant sex differences in the overall discriminative ability, as measured by the AUC, for any plasma biomarker in either the CU or CI groups (all DeLong test p > 0.05; Figure 2 and Tables S3 and S4 in supporting information).

FIGURE 2.

Core Alzheimer's disease plasma biomarkers’ performance to identify cognitively unimpaired and cognitively impaired amyloid PET‐positive individuals in the overall cohort and by sex. ROC analyses were conducted for the discrimination between amyloid PET‐positive and amyloid PET‐negative individuals in the entire sample and stratified by sex. AUC differences were tested using a two‐sided DeLong test. We determined the optimal threshold for the entire sample using the Youden index. To assess performance consistency, this common threshold was applied to men and women separately. At this threshold, we calculated accuracy, sensitivity, specificity, PPV, and NPV, with 95% CIs derived from 1000 bootstrap resamples. The error bars denote the 95% CIs. Sex differences in accuracy, sensitivity, and specificity were evaluated using the Pearson χ ² test. PPV and NPV were compared using permutation tests (10,000 iterations). Amyloid PET positivity was established at 25 Centiloid. *p < 0.05 compared to men; **p < 0.01 compared to men. Aβ, amyloid beta; AUC, area under the curve; CI confidence interval; IQR, interquartile range; NPV, negative predictive value; PET, positron emission tomography; PPV, positive predictive value; p‐tau, phosphorylated tau; ROC, receiver operating characteristic

However, when applying the common, cohort‐derived optimal thresholds, we observed sex differences in specific classification metrics: among CU individuals, Fujirebio p‐tau217 demonstrated higher accuracy in women (80.8% in women vs. 71.6% in men; p = 0.011), which was driven by a higher specificity (80.8% in women vs. 66.3% in men; p = 0.001) and PPV (66.2% in women vs. 52.8% in men; p = 0.033), with comparable sensitivity (p = 0.79) and NPV (p = 0.85). This pattern was mirrored by the p‐tau217/Aβ42 ratio, which also showed significantly higher accuracy (83.1% vs. 76.0%; p = 0.033) and specificity (82.9% vs. 70.9%; p = 0.006) in women compared to men, with comparable sensitivity (p = 0.62; Figure 2; Table S3). In contrast, among CI participants, the Fujirebio p‐tau217/Aβ42 ratio showed significantly lower specificity in women compared to men (87.3% vs. 95.7%; p = 0.042). No other significant sex‐based differences in classification performance were observed for any biomarker in this group (Figure 2; Table S4 in supporting information).

We next derived and applied sex‐specific optimal thresholds to determine whether this approach would mitigate the performance differences observed with the common cutoff (Tables S5 and S6 in supporting information). Within CU participants, the pattern of higher specificity and PPV in women for the Fujirebio assays remained such that Fujirebio p‐tau217 had a significantly higher specificity in women (86.8% in women vs. 59.3% in men, p < 0.0001), as was PPV (73.3% in women vs. 50.7% in men, p = 0.0004). Notably, this was accompanied by lower sensitivity in women compared to men (78.0% vs. 92.3%, p = 0.015). Similarly, Fujirebio p‐tau217/Aβ42 ratio showed a higher accuracy (84.3% vs. 75.6%, p = 0.001), specificity (85.5% vs. 69.8%, p = 0.0002), and PPV (72.4% vs. 57.0%, p = 0.009) in women than in men (Table S5). Within CI participants, several biomarkers demonstrated lower performance in women. The Fujirebio p‐tau217 and p‐tau217/Aβ42 ratio had significantly less specificity in women than in men (84.3% vs. 94.0%, p = 0.034; and 87.3% vs. 95.7%, p = 0.042, respectively). The C2N p‐tau217 assay was significantly less accurate (67.3% vs. 86.8%, p = 0.006), less sensitive (60.5% vs. 85.1%, p = 0.002), and had a significantly lower NPV (37.0% vs. 68.8%, p = 0.003) in women compared to men (Table S6).

Next, we evaluated performance using a two‐cutoff approach, in which thresholds for positivity and negativity were derived from the entire cohort and then applied to men and women separately (Tables S7 and S8 in supporting information). Within CU participants, this analysis showed a similar pattern to the single‐cutoff approach. The Fujirebio p‐tau217/Aβ42 ratio demonstrated higher accuracy (88.9% vs. 82.6%, p = 0.022), specificity (91.8% vs. 80.5%, p = 0.006), and PPV (81.0% vs. 63.8%, p = 0.026) in women compared to men, though the differences were not statistically significant for p‐tau217 alone (Table S7). In CI participants (Table S8), the Fujirebio p‐tau217/Aβ42 ratio had a significantly lower specificity in women than in men (86.1% vs. 94.8%, p = 0.048). No other statistically significant sex differences were observed. The proportion of cases classified as intermediate did not significantly differ between sexes in either cognitive group.

Finally, we derived and applied two‐cutoff models that were optimized separately for each sex (Tables S9 and S10 in supporting information). There were no significant sex differences in classification performance parameters in CU (Table S9) or CI groups (Table S10). However, a significantly higher proportion of women were classified as “intermediate” compared to men for several key biomarkers. Specifically, in CU individuals, this was observed for the C2N Aβ42/40 ratio (63.8% in women vs. 29.9% in men, p < 0.0001). In CI individuals, a higher proportion of women were classified as intermediate for the Fujirebio Aβ42/40 (53.8% in women vs. 43.6% in men, p = 0.008) and p‐tau217/Aβ42 (5.2% in women vs. 0.5% in men, p < 0.0001), as well as for C2N p‐tau217 (35.3% in women vs. 6.4% in men, p < 0.0001).

3.3. Sex differences in the association of plasma biomarker–based amyloid PET positivity with cognitive trajectories

For individuals with available plasma biomarkers and subsequent longitudinal cognitive data, we tested whether a positive core AD plasma biomarker status indicative of amyloid PET positivity (plasma Aβ42/40, p‐tau217, and p‐tau217/Aβ42) was associated with faster rates of cognitive decline and whether this association was modified by sex. For reference, the analysis was also performed using amyloid PET positivity stratification. Characteristics of participants with longitudinal cognitive data at their first plasma visit are shown in Tables S11 and S12 in supporting information. Among the 504 CU individuals with Fujirebio plasma biomarkers, 101 (20.0%) progressed to CI by their last follow‐up visit, including 55 men (25.2%) and 46 women (16.1%). Among the 222 CU individuals with C2N plasma biomarkers, 39 (17.6%) progressed to CI including 18 men (18.0%) and 21 women (17.2%).

In the whole CU sample, a positive status for amyloid PET, the Fujirebio p‐tau217/Aβ42 ratio, and Fujirebio and C2N p‐tau217 in plasma were associated with a greater decline on the mPACC over time (Table S13 in supporting information). Crucially, we found a significant three‐way interaction among biomarker status, sex, and time for both Fujirebio p‐tau217 (p = 0.003) and C2N p‐tau217 (p = 0.004), and the Fujirebio p‐tau217/Aβ42 ratio (p = 0.0009; Figure 3 and Table S13). Post hoc sex‐stratified analyses showed that a positive status on these plasma biomarkers was associated with a decline only in women (Fujirebio p‐tau217: β = −0.72, p < 0.0001 in women and β = −0.18, p = 0.23 in men; C2N p‐tau217: β = −0.63, p = 0.0006 in women and β = 0.17, p = 0.39 in men; p‐tau217/Aβ42 ratio: β = −0.79, p < 0.0001 in women and β = −0.15, p = 0.31 in men; Table S13). Similarly, amyloid PET positivity was also significantly associated with mPACC decline in women (β = −0.92, p < 0.0001) and not in men (β = −0.49, p = 0.17) in stratified analyses, although the interaction term was not significant (p = 0.12). In sensitivity analyses including GFAP as a covariate, the sex interaction remained statistically significant for Fujirebio p‐tau217 and p‐tau217/Aβ42 (β = −0.59, p = 0.002 and β = −0.64, p = 0.0006, respectively). These analyses could not be performed for C2N p‐tau217 due to the low number (n = 32) of CU individuals with concurrent Quanterix biomarkers. Similarly, further adjusting the models for BMI did not alter these results (β = −0.58, p = 0.002 for Fujirebio p‐tau217; β = −0.63, p = 0.0009 for Fujirebio p‐tau217/Aβ42; β = −0.81, p = 0.002 for C2N p‐tau217).

FIGURE 3.

Association of core Alzheimer's disease plasma biomarker status with cognitive trajectories by sex. Longitudinal trajectories of mPACC scores (for CU individuals) or CDR‐SB (for CI individuals) by sex and plasma biomarker status. Each line connects multiple observations from the same individual. Participants were classified as plasma biomarker positive or negative using binary cutoffs derived from the entire sample. Linear mixed effects models with random intercepts and slopes were fitted. p values for the three‐way interaction among sex, time, and biomarker group are shown. Models were adjusted for the interactions of time with age, APOE ε4 status, and education attainment. Asterisks (*) indicate statistical significance after FDR correction for multiple comparisons. AD, Alzheimer's disease; APOE, apolipoprotein E; CDR‐SB, Clinical Dementia Rating Sum of Boxes; CI, cognitively impaired; CU, cognitively unimpaired; FDR, false discovery rate; mPACC, modified Preclinical Alzheimer Cognitive Composite

In the whole CI sample (Table S14 in supporting information), a positive status for amyloid PET and for all plasma biomarkers was associated with a steeper decline on the CDR‐SB over time. Unlike in the CU group, there were no significant interactions with sex (all p > 0.05), with biomarkers being associated with cognitive decline similarly in men and women in the CI stage (Figure 3 and Table S14). Using ADAS‐Cog 13 as cognitive outcome in the CI group rendered very similar results (Table S15 in supporting information). Finally, sensitivity analyses restricted to participants with MCI (CDR = 0.5) yielded results consistent with the primary analyses, confirming that findings in the CI group were not driven by the inclusion of individuals with dementia (Tables S16 and S17 in supporting information).

4. DISCUSSION

In this study, we assessed sex differences in AD plasma biomarker levels, and their clinical utility in CU and CI ADNI individuals. Our main findings were: (1) plasma GFAP was higher in women across both CU and CI individuals. In the CU group, men showed lower Aβ42 and higher p‐tau217/Aβ42 ratio, whereas in CI individuals, p‐tau217 and p‐tau217/Aβ42 were higher in women. (2) Despite similar overall classification performance, classification metrics differed by sex particularly for p‐tau217 and p‐tau217/Aβ42, and differences depended on clinical disease stage. (3) In CU individuals, plasma p‐tau217 and p‐tau217/Aβ42‐based amyloid positivity was predictive of cognitive decline only in women, while no sex differences in the plasma biomarkers’ association with cognitive decline were observed in CI individuals.

Our study revealed significant sex differences in plasma biomarker levels. Plasma GFAP concentrations were consistently higher in women than in men across both CU and CI groups. This finding aligns with previous reports ¹³ , ¹⁵ , ⁴⁴ and may reflect sex differences in astrocytic responses to amyloid pathology, which could influence downstream tau accumulation. ⁶⁵ , ⁶⁶ , ⁶⁷ Other differences in plasma biomarker levels were found that depended on disease stage. In CU individuals, both Fujirebio and C2N plasma Aβ42 were lower in men; however, the absence of a lower Aβ42/40 ratio or higher amyloid PET burden in men suggests that this difference may reflect variability in Aβ42 production, clearance, or on peripheral factors influencing Aβ42 plasma levels rather than amyloid deposition. ⁵⁰ , ⁶⁸ Notably, because p‐tau217 did not differ by sex in CU individuals, the higher Fujirebio p‐tau217/Aβ42 ratio observed in men was likely driven by lower Aβ42. In CI individuals, all plasma p‐tau217 measures were higher in women, a finding that is consistent with reports of greater tau pathology burden and more advanced disease stage in women compared to men. ²⁴ , ²⁵ , ²⁶ , ²⁷ , ³⁰ , ⁶⁹ This pattern is further supported by the higher mesial‐temporal tau PET signal observed in women across both cognitive groups. However, among CU individuals, this higher regional tau burden did not yet appear to be reflected in plasma biomarker levels, which may be explained by plasma p‐tau217 being more tightly linked to amyloid than tau pathology in CU individuals. ⁷⁰ ,  ,  ,  

Observed sex differences in plasma biomarker levels were not explained by differences in BMI or in prevalence of comorbidities, ³⁷ , ⁴⁸ , ⁴⁹ , ⁷¹ suggesting the presence of intrinsic biological differences between sexes for these biomarkers. Furthermore, sex‐related differences in blood–brain barrier (BBB) integrity or transport mechanisms may also contribute to observed variations in plasma biomarker levels. Although direct measures of BBB function were not available in this study, future research should assess its role as a potential biological modifier of plasma biomarker concentrations.

Next, we evaluated sex differences in the performance of core AD plasma biomarkers to identify amyloid PET positivity, a critical question for their use in clinical trials. While the overall classification performance (AUC) did not differ between sexes, key performance metrics varied by sex and disease stage. Specifically, for Fujirebio p‐tau217 and its ratio with Aβ42, classification accuracy was higher in CU women (driven by greater specificity and PPV), whereas the opposite was true in CI individuals, in which men showed higher specificity. These differences persisted even after applying sex‐optimized cutoffs, suggesting they are not fully explained by differences in biomarker levels and may reflect more complex biological interactions. Although a two‐cutoff approach attenuated these differences, it resulted in a larger “intermediate” range for women, pointing to greater biological variability or overlap in biomarker distributions. Overall, while sex‐specific cutoffs may not be necessary, these findings have important implications. Using a unified cutoff in prevention trials may lead to a higher rate of false‐positive classifications in CU men; however, in trials for symptomatic individuals, the same approach could result in more false positives among women, potentially impacting the assessment of treatment efficacy.

Finally, we examined whether the association between plasma biomarker–based amyloid positivity stratification and rates of cognitive decline differed by sex. We applied uniform thresholds for this analysis because our data‐driven derivation of sex‐specific cutoffs did not indicate a consistently higher threshold was necessary for men or women. Furthermore, applying arbitrarily different thresholds by sex would restrict the sample to those with the highest pathological burden, potentially confounding the analysis by selecting for disease severity rather than sex‐specific biology. Among CU individuals, a positive biomarker status for Fujirebio p‐tau217, p‐tau217/Aβ42, and C2N p‐tau217 was associated with faster decline in mPACC over time, specifically in women. A similar trend was observed for amyloid PET status, although the three‐way interaction among sex, time, and amyloid PET status did not reach statistical significance. These findings suggest greater vulnerability to downstream effects of AD pathology, particularly p‐tau–related processes, even at preclinical stages. The higher tau PET burden observed in CU women may further contribute to this faster cognitive decline. It is worth noting that while higher BMI in late life has been associated with paradoxical protective effects on cognition, particularly in women, ⁶³ BMI did not differ between sexes in our CU sample (p = 0.72). Furthermore, sensitivity analyses confirmed that the sex‐specific prognostic value of plasma biomarkers remained robust after adjusting for BMI or baseline plasma GFAP concentration, indicating that neither the BMI paradox nor higher GFAP levels in women are likely to drive our findings. Future studies with larger sample sizes could further explore potential synergistic interactions among amyloid, tau, and astrocytic biomarkers in a sex‐dependent manner. In contrast, in CI individuals, a positive plasma biomarker status was associated with a similar rate of cognitive decline in women and men, despite higher p‐tau levels in women. This pattern suggests that once a certain pathological burden is reached, cognitive decline progresses at a comparable rate across sexes.

Remarkably, results were consistent across CDR‐SB and ADAS‐Cog13, the most widely used cognitive outcomes in clinical trials. The fact that these results were observed for p‐tau217 biomarkers and not for Aβ42/40 is consistent with previous literature supporting a tighter association of tau biomarkers with cognitive decline, ⁷² , ⁷³ , ⁷⁴ , ⁷⁵ as well as higher tau vulnerability among women. ²² , ²³ , ²⁴ , ²⁵ Consistent with our findings, previous studies on tau PET have reported stronger tau‐related cognitive decline in CU women, ²⁴ , ⁷⁶ and a recent study found a female‐specific association between plasma p‐tau217 and cognitive and brain atrophy trajectories specifically in CU individuals. ⁴² Our results extend these observations in a larger ADNI sample and across multiple AD plasma biomarkers.

Strengths of our study include being the first to comprehensively examine sex differences in key AD plasma biomarkers and their impact on biomarker performance in clinical trial contexts, leveraging the large and well‐characterized ADNI cohort. Inclusion of two major analytical platforms (Fujirebio and C2N) adds robustness and generalizability to our findings. While sample sizes for C2N biomarkers were smaller than for Fujirebio, the strong agreement observed between these platforms in our study and others ⁵ supports the robustness of our sex‐specific findings across different assays. Moreover, we applied a comprehensive analytical framework to assess sex differences in classification performance beyond simple stratification, by testing whether a unified cutoff performed differently by sex and by further evaluating sex‐optimized models. However, our study has several limitations that should be acknowledged. First, although we did adjust our analyses for BMI or comorbidities when sex differences were present, information on certain comorbid conditions such as diabetes, dyslipidemia, and chronic kidney disease was obtained at study entry rather than at the time of plasma biomarker collection. Due to the lack of concurrent data at plasma collection, we could not explore additional peripheral factors potentially influencing plasma levels, such as kidney function. Second, plasma biomarker cutoffs were optimized for amyloid PET positivity, introducing a degree of circularity in the classification accuracy analyses. Third, mPACC and ADAS‐Cog are heavily weighted toward verbal memory measures. Given the well‐established female advantage in verbal domains, they may effectively mask early cognitive deficits. ⁷⁷ Consequently, the steeper decline observed in biomarker‐positive CU women might reflect a rapid deterioration occurring once this verbal cognitive reserve is exhausted. Future studies incorporating non‐verbal memory measures are needed to fully disentangle these domain‐specific sex differences.

In summary, we comprehensively studied potential sex differences in the levels and utility of main AD plasma biomarkers and found that they differ by sex and disease stage, particularly for p‐tau217 or p‐tau217/Aβ42. Although our findings indicate that implementing sex‐specific biomarker cutoffs does not significantly improve identification of amyloid PET positivity, sex remains a critical biological variable. Among CU individuals, higher specificity and PPV in women, together with the stronger association between biomarker positivity and subsequent cognitive decline, suggest that women may be more reliably identified as at risk of AD using current markers. In contrast, among CI participants, amyloid PET positivity classification accuracy appeared slightly higher in men, and biomarker positivity predicted cognitive decline similarly in men and women. These results highlight that while diagnostic thresholds may remain uniform, consideration of sex differences, particularly in preclinical and prodromal populations, is essential to optimize trial recruitment, improve sensitivity to treatment effects, and enhance clinical interpretation of plasma biomarker results.

CONFLICT OF INTEREST STATEMENT

M.M.A. reports support for the present manuscript from the Alzheimer's Association Research Fellowship (AARF‐23‐1141384). I.H. reports support for the present manuscript from the National Institute on Aging. A.M., C.S., and ADNI report no conflicts of interest. P.T. reports support for the present manuscript via an NIH grant to their institution. L.M.S. reports grants from NIA/NIH (P30 AG072979; U19 AG024904; R01 AG06) and the Department of Defense (W81XWH2211081‐B) paid to their institution. L.M.S. also reports consulting fees and honoraria for lectures from Biogen and Roche, as well as in‐kind support (reagents/equipment) from Fujirebio and Roche for the ADNI4 study paid to their institution. M.W.W. reports institutional research support from NIH, Department of Defense, California Department of Public Health, Siemens, Biogen, Hillblom Foundation, Alzheimer's Association, Johnson & Johnson, Kevin and Connie Shanahan, GE, VUmc, Australian Catholic University, The Stroke Foundation, and the Veterans Administration. M.W.W. reports consulting fees from Acadia Pharmaceuticals, Acumen Pharmaceuticals, Boxer Capital, BrightFocus Foundation, Cerecin, Clario/BioClinica, Dementia Society of Japan, Eisai, Guidepoint, Health and Wellness Partners, Indiana University, LCN Consulting, Merck Sharp & Dohme Corp., Duke University, Owkin France, NovoNordisk, ProMIS Neurosciences, Prova Education, Sai Med Partners, T3D Therapeutics, USC, WebMD, MEDA Corp., Quantum Leap Health, REGENLIFE, GLG Consulting, and IXICO. He reports honoraria and/or travel support from the China Association for Alzheimer's Disease, Taipei Medical University, Cleveland Clinic, Banner Health, AD/PD Congress, Foundation of Learning, Health Society (Japan), INSPIRE Project (U. Toulouse), Japan Society for Dementia Research, Korean Dementia Society, National Center for Geriatrics and Gerontology (Japan), and University of Wisconsin Madison. He serves on advisory boards/committees for ADNI, UCSF, ProMIS Neurosciences, Acumen Pharmaceuticals, and Duke University. He holds stock/options in Alzeca, Alzheon, Inc., ALZPath, and Anven. D.T. reports NIH grants paid to their institution; grants from the Michael J. Fox Foundation (MJFF); consulting fees from Alzheon, Veravas, and Roche; and honoraria from NIH, MJFF, and Aligning Science Across Parkinson's (ASAP). Author disclosures are available in the supporting information.

CONSENT STATEMENT

Written informed consent was obtained from each participant or their legally authorized representative.

Supporting information

Supporting Information

Supporting Information