diff --git a/ace/scrape.py b/ace/scrape.py index 00ca9c2..206e4e9 100644 --- a/ace/scrape.py +++ b/ace/scrape.py @@ -235,33 +235,101 @@ def parse_PMID_xml(xml): return metadata +# Substrings that only ever show up on a blocked/failed scrape. +BLOCKED_PAGE_PATTERNS = ['Checking if you are a human', +'Please turn JavaScript on and reload the page', +'Checking if the site connection is secure', +'Enable JavaScript and cookies to continue', +'There was a problem providing the content you requested', +'Redirecting', +'Page not available - PMC', +'Your request cannot be processed at this time. Please try again later', +'403 Forbidden', +'Page not found — ScienceDirect', +'This site can’t be reached', +'used Cloudflare to restrict access', +'502 Bad Gateway', +'Checking your browser before accessing', +'Checking your browser - reCAPTCHA', +'/recaptcha/challengepage/', +] + +# Markers for an embedded reCAPTCHA widget. These are *not* proof of a block: +# PubMed, PMC and several publishers put one in ordinary page furniture (the +# "Email" form on a PubMed abstract page, for instance). They only indicate a +# challenge page when the widget is essentially all the page contains. +RECAPTCHA_PATTERNS = ['g-recaptcha'] + +# A real article page renders far more text than a challenge interstitial, +# which only says something like "Checking your browser before accessing ...". +MIN_VISIBLE_CHARS = 1000 + + +# PubMed journal names of the PLoS titles, mapped to their journals.plos.org +# site slug. PLoS serves NLM XML (with real, parseable tables) for all of them. +PLOS_JOURNAL_SITES = { + 'plos one': 'plosone', + 'plos biology': 'plosbiology', + 'plos medicine': 'plosmedicine', + 'plos computational biology': 'ploscompbiol', + 'plos genetics': 'plosgenetics', + 'plos pathogens': 'plospathogens', + 'plos neglected tropical diseases': 'plosntds', +} + + +def _is_pubmed_landing_page(url): + """ True for the PubMed abstract page, which is where prlinks drops us when + an article has no LinkOut record. It is not full text. """ + + return re.match(r'https?://pubmed\.ncbi\.nlm\.nih\.gov/\d+/?$', url or '') is not None + + +def _extract_citation_doi(html): + """ Pull the DOI out of a page's citation_doi meta tag. """ + + if not html: + return None + + for tag in re.findall(r'(?is)]*>', html): + if not re.search(r'''(?i)name\s*=\s*["']?citation_doi["']?''', tag): + continue + match = re.search(r'''(?i)content\s*=\s*["']([^"']+)["']''', tag) + if match: + return match.group(1).strip() + + return None + + +def _plos_doi_from_url(url): + """ Pull the DOI out of a journals.plos.org article URL, if it is one. """ + + match = re.search(r'article\?id=([^&#]+)', url or '') + return match.group(1) if match else None + + +def _visible_text(html): + """ Rough plain-text rendering of an HTML document, used to tell a real + article page apart from an interstitial that carries nothing but a widget. """ + + text = re.sub(r'(?is)<(script|style|noscript)\b[^>]*>.*?', ' ', html) + text = re.sub(r'(?s)', ' ', text) + text = re.sub(r'(?s)<[^>]*>', ' ', text) + return re.sub(r'\s+', ' ', text).strip() + + def _validate_scrape(html): - """ Checks to see if scraping was successful. + """ Checks to see if scraping was successful. For example, checks to see if Cloudfare interfered """ - patterns = ['Checking if you are a human', - 'Please turn JavaScript on and reload the page', - 'Checking if the site connection is secure', - 'Enable JavaScript and cookies to continue', - 'There was a problem providing the content you requested', - 'Redirecting', - 'Page not available - PMC', - 'Your request cannot be processed at this time. Please try again later', - '403 Forbidden', - 'Page not found — ScienceDirect', - 'This site can’t be reached', - 'used Cloudflare to restrict access', - '502 Bad Gateway', - 'Checking your browser before accessing', - 'Checking your browser - reCAPTCHA', - '/recaptcha/challengepage/', - 'g-recaptcha', - ] - - for pattern in patterns: + for pattern in BLOCKED_PAGE_PATTERNS: if pattern in html: return False + if any(pattern in html for pattern in RECAPTCHA_PATTERNS): + if len(_visible_text(html)) < MIN_VISIBLE_CHARS: + return False + return True ''' Class for journal Scraping. The above free-floating methods should @@ -639,9 +707,19 @@ def check_for_substitute_url(self, url, html, journal): j = journal.lower() try: - if j == 'plos one': - doi_part = re.search('article\?id\=(.*)', url).group(1) - return 'http://journals.plos.org/plosone/article/asset?id=%s.XML' % doi_part + if j in PLOS_JOURNAL_SITES: + doi = _plos_doi_from_url(url) + if doi is None and _is_pubmed_landing_page(url): + # E-utilities had no LinkOut for this article, so prlinks + # bounced us to the PubMed abstract page. Its citation_doi + # meta tag is enough to address the full text ourselves. + doi = _extract_citation_doi(html) + if doi is None: + return url + # The legacy 'asset?id=.XML' endpoint still works, but only + # by redirecting here first. + return 'https://journals.plos.org/%s/article/file?id=%s&type=manuscript' % ( + PLOS_JOURNAL_SITES[j], doi) elif j in ['human brain mapping', 'european journal of neuroscience', 'brain and behavior', 'epilepsia', 'journal of neuroimaging']: return url.replace('abstract', 'full').split(';')[0] diff --git a/ace/tests/cassettes/test_ace/test_journal_scraping.yaml b/ace/tests/cassettes/test_ace/test_journal_scraping.yaml index f23653c..ed6ffec 100644 --- a/ace/tests/cassettes/test_ace/test_journal_scraping.yaml +++ b/ace/tests/cassettes/test_ace/test_journal_scraping.yaml @@ -493,8 +493,6 @@ interactions: May 2025 23:51:37 GMT Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked X-RateLimit-Limit: - '3' X-RateLimit-Remaining: @@ -609,8 +607,6 @@ interactions: May 2025 23:51:39 GMT Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked X-RateLimit-Limit: - '3' X-RateLimit-Remaining: @@ -725,8 +721,6 @@ interactions: May 2025 23:51:39 GMT Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked X-RateLimit-Limit: - '3' X-RateLimit-Remaining: @@ -780,8 +774,6 @@ interactions: 23:51:39 GMT; HttpOnly; Max-Age=31536000; Path=/; Secure Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked Vary: - Cookie,origin Via: @@ -1213,8 +1205,6 @@ interactions: 23:51:40 GMT; HttpOnly; Max-Age=31536000; Path=/; Secure Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked Vary: - origin Via: @@ -1333,8 +1323,6 @@ interactions: May 2025 23:51:42 GMT Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked X-RateLimit-Limit: - '3' X-RateLimit-Remaining: @@ -1449,8 +1437,6 @@ interactions: May 2025 23:51:42 GMT Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked X-RateLimit-Limit: - '3' X-RateLimit-Remaining: @@ -1504,8 +1490,6 @@ interactions: 23:51:42 GMT; HttpOnly; Max-Age=31536000; Path=/; Secure Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked Vary: - Cookie,origin Via: @@ -1974,8 +1958,6 @@ interactions: 23:51:42 GMT; HttpOnly; Max-Age=31536000; Path=/; Secure Strict-Transport-Security: - max-age=31536000; includeSubDomains; preload - Transfer-Encoding: - - chunked Vary: - origin Via: @@ -4053,4 +4035,3601 @@ interactions: status: code: 200 message: OK +- request: + body: null + headers: + Accept: + - '*/*' + Accept-Encoding: + - gzip, deflate + Connection: + - keep-alive + User-Agent: + - Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like + Gecko) Chrome/123.0.6312.58 Safari/537.36 + method: GET + uri: https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0300720&type=manuscript + response: + body: + string: '' + headers: + Alt-Svc: + - h3=":443"; ma=2592000,h3-29=":443"; ma=2592000 + Cache-Control: + - max-age=3600 + Content-Length: + - '0' + Date: + - Wed, 23 Sep 2026 16:31:40 GMT + Location: + - https://storage.googleapis.com/plos-corpus-prod/10.1371/journal.pone.0300720/1/pone.0300720.xml?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=wombat-sa%40plos-prod.iam.gserviceaccount.com%2F20260923%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20260923T163009Z&X-Goog-Expires=86400&X-Goog-SignedHeaders=host&X-Goog-Signature=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 + Server: + - Apache + Set-Cookie: + - GCLB=CMK6yPmI7J_noAEQAw; path=/; HttpOnly + Strict-Transport-Security: + - max-age=31536000 + - max-age=31536000 ; includeSubDomains + Via: + - 1.1 google + X-Content-Type-Options: + - nosniff + X-Frame-Options: + - DENY + X-XSS-Protection: + - 1; mode=block + status: + code: 302 + message: Found +- request: + body: null + headers: + Accept: + - '*/*' + Accept-Encoding: + - gzip, deflate + Connection: + - keep-alive + User-Agent: + - Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like + Gecko) Chrome/123.0.6312.58 Safari/537.36 + method: GET + uri: https://storage.googleapis.com/plos-corpus-prod/10.1371/journal.pone.0300720/1/pone.0300720.xml?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=wombat-sa%40plos-prod.iam.gserviceaccount.com%2F20260923%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20260923T163009Z&X-Goog-Expires=86400&X-Goog-SignedHeaders=host&X-Goog-Signature=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 + response: + body: + string: "\n\n
\n\n\nPLoS ONE\nplos\nplosone\n\nPLOS + ONE\n\n1932-6203\n\nPublic + Library of Science\nSan Francisco, CA USA\n\n\n\n10.1371/journal.pone.0300720\nPONE-D-23-39384\n\n\nResearch Article\n\n\nPeople and placesPopulation + groupingsAge groups\nBiology and life sciencesDevelopmental + biologyOrganism developmentAging\nBiology and life sciencesPhysiologyPhysiological + processesAging\nBiology and life sciencesAnatomyHeadEyes\nMedicine and health sciencesAnatomyHeadEyes\nBiology and life sciencesAnatomyOcular + systemEyes\nMedicine and health sciencesAnatomyOcular + systemEyes\nBiology and life sciencesNeuroscienceBrain + mappingFunctional magnetic resonance imagingResting + state functional magnetic resonance imaging\nMedicine and health sciencesDiagnostic + medicineDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imagingResting state functional magnetic resonance + imaging\nResearch and analysis methodsImaging + techniquesDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imagingResting state functional magnetic resonance + imaging\nMedicine and health sciencesRadiology + and imagingDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imagingResting state functional magnetic resonance + imaging\nResearch and analysis methodsImaging + techniquesNeuroimagingFunctional + magnetic resonance imagingResting state functional + magnetic resonance imaging\nBiology and life sciencesNeuroscienceNeuroimagingFunctional + magnetic resonance imagingResting state functional + magnetic resonance imaging\nBiology and life sciencesNeuroscienceBrain + mappingFunctional magnetic resonance imaging\nMedicine and health sciencesDiagnostic + medicineDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imaging\nResearch and analysis methodsImaging + techniquesDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imaging\nMedicine and health sciencesRadiology + and imagingDiagnostic radiologyMagnetic + resonance imagingFunctional magnetic resonance + imaging\nResearch and analysis methodsImaging + techniquesNeuroimagingFunctional + magnetic resonance imaging\nBiology and life sciencesNeuroscienceNeuroimagingFunctional + magnetic resonance imaging\nComputer and information sciencesArtificial + intelligenceMachine learningSupport + vector machines\nResearch and analysis methodsMathematical + and statistical techniquesStatistical methodsRegression + analysisLinear regression analysis\nPhysical sciencesMathematicsStatisticsStatistical + methodsRegression analysisLinear + regression analysis\nComputer and information sciencesSoftware + engineeringPreprocessing\nEngineering and technologySoftware + engineeringPreprocessing\n\nBrain + age monotonicity and functional connectivity differences of healthy subjects\nBrain age monotonicity and functional connectivity + differences\n\n\n\nhttps://orcid.org/0000-0002-1172-6291\n\nSorooshyari\nSiamak + K.\n\nConceptualization\nFormal + analysis\nWriting + \u2013 original draft\n*\n\n\n\nDepartment + of Statistics, Stanford University, Stanford, CA, United States of America\n\n\n\nHajebrahimi\nFarzin\n\nEditor\n\n\n\nNew + Jersey Institute of Technology, UNITED STATES\n\n\n

The authors have declared that no + competing interests exist.

\n\n* E-mail: siamak@stanford.edu\n\n\n30\n5\n2024\n\n\n2024\n\n19\n5\ne0300720\n\n\n27\n11\n2023\n\n\n4\n3\n2024\n\n\n\n2024\nSiamak + K. Sorooshyari\n\nThis is an open access article distributed + under the terms of the Creative Commons Attribution License, which + permits unrestricted use, distribution, and reproduction in any medium, provided + the original author and source are credited.\n\n\n\n\n

Alterations + in the brain\u2019s connectivity or the interactions among brain regions have + been studied with the aid of resting state (rs)fMRI data attained from large + numbers of healthy subjects of various demographics. This has been instrumental + in providing insight into how a phenotype as fundamental as age affects the + brain. Although machine learning (ML) techniques have already been deployed + in such studies, novel questions are investigated in this work. We study whether + young brains develop properties that progressively resemble those of aged + brains, and if the aging dynamics of older brains provide information about + the aging trajectory in young subjects. The degree of a prospective monotonic + relationship will be quantified, and hypotheses of brain aging trajectories + will be tested via ML. Furthermore, the degree of functional connectivity + across the age spectrum of three datasets will be compared at a population + level and across sexes. The findings scrutinize similarities and differences + among the male and female subjects at greater detail than previously performed.

\n\n\nThe + author(s) received no specific funding for this work.\n\n\n\n\n\n\n\n\nData Availability\nThe + data underlying the results presented in the study are available from 1000FCP + (http://fcon_1000.projects.nitrc.org), camCAN + (https://www.cam-can.org/index.php?content=dataset), + and SRPBS (https://bicr-resource.atr.jp/srpbsopen). + The 1000FCP and camCAN datasets were further processed to arrive at the correlations + that were used as features in the analysis. The processing is outlined in + the Materials and Methods of the manuscript, and the processed correlations. + The code for the data processing and ML analysis, as well as the processed + data are available from the Github repository (https://github.com/sorooshyari/PLOS_ONE_2004).\n\n\n\n\n\n\nIntroduction\n

An understanding + of the changes that occur to a brain is fundamental to human neuroscience. + While variability in person-to-person brain aging is expected due to genetics, + environment, and life events, there are also changes that are more salient. + Resting-state fMRI can be collected relatively quickly and easily from subjects + of different ages, demographics, and possible pathologies. Atypical brain + aging is often reflective of or a precursor to developmental disorders. Furthermore, + abnormal trajectories in brain age have been implicated in Parkinson\u2019s + disease (PD) [1], + Alzheimer\u2019s disease (AD) [2], + as well as several other neural pathologies [3] + and even mortality [4]. + Utilizing fMRI data to investigate brain aging dynamics entails examining + alterations in measures of functional connectivity (FC). Typically, patterns + of correlated activity between regions of interest (ROIs) are compared between + conditions to assess whether there are changes. In the case of aging, differences + between connectivity maps are quantified and used to predict age with the + purpose of identifying if and how brain signals change over time. Due to its + importance, age prediction from brain activity has been undertaken across + sizable datasets.

\n

Subsequent to correlational studies and graph-theoretic + approaches, machine learning (ML) has been used to discover structure by performing + either regression or classification. There exists a growing number of works + that use ML to answer questions pertaining to brain aging. In [5] support vector regression (SVR) was + used to estimate brain age. Three groups of subjects were considered with + the ages of a subject lying in between 6 and 35 years. The considered age + range was limited in comparison to more recent works as well as the data considered + in this work. The work considered 12,720 FC features per subject and pare + to 200 for the analysis. LASSO regression was used in [6] to predict brain age by considering + over 14,000 healthy participants from the UK Biobank. Interestingly, that + work found that resting and task-based fMRI are relatively uninformative in + comparison to several variants of MRI, i.e. T1-weighted MRI, T2-FLAIR,T2*, + and diffusion-MRI, at chronological age prediction. A review on the use of + ML techniques for rsfMRI analysis is provided in [7]. We shall highlight several of the more + recent and particularly relevant works below.

\n

The deviation of a person\u2019s + brain age from biological age may also be affected by sex as there is no reason + to assume that the prospective changes are uniform among males and females. + Works are increasingly incorporating sex as either a predictor or a response + variable when evaluating brain age. A recent study [8] considered over 4,000 subjects across + a wide age spectrum from five different sites. Ridge regression was used, + and brain age prediction performance was quantified via the mean absolute + error (MAE) and correlation between the predicted and actual ages. The work + quantified the effects of a multiscale study by considering various numbers + of functional networks in the predictive analysis. In [9] the authors consider over 9,000 different + pipelines consisting of hyperparameter values across a large (over 14,000) + number of subjects from the Human Connectome Project (HCP) and UK Biobank. + Correlation-based measures of functional connectivity were computed and used + as inputs for several classifiers including variants of deep learning (DL). + It was noted that the regularized regression method elastic net performs comparably + with DL schemes for age prediction.

\n

We use an ML pipeline to study + whether brain aging is a progressive phenomenon that can be characterized + by a monotonic relationship. Recordings of different age groups will be used + in a training procedure with the resulting machine asked to classify subjects + from age groups that the machine has not previously encountered. The findings + will reveal a prospective similarity or difference among the connectivity + of brains from a test set to those that are younger or older. Several hypotheses + will be presented and tested. Namely, what degree of monotonicity exists in + the changes experienced by brains during aging? How consistent are the findings + across datasets? Does the aging trajectory of young brains provide insight + into the aging trajectory of older brains? Conversely, are alterations in + older subjects informative of brain activity earlier in life? To the best + of our knowledge, such analysis has not been previously undertaken in imaging + studies that assess FC.

\n

The dedifferentiation theory of aging was + presented to reflect evidence of brain activity that is localized to specific + regions in young adults becoming less localized with aging [10]. While there exist advancements to + and affirmations of the dedifferentiation theory of aging [11, 12], + we scrutinize the variability in the FC that appears at the various age groups + by computing the number of significant connections across the brains of subjects + to evaluate how they are altered by aging. The study shall be undertaken separately + for male and female subjects as well to assess sex differences. The analysis + constitutes novel findings and quantifies the modularity with the inclusion + of anticorrelations. The majority of FC fMRI works restrict attention to positive + connections and do not differentiate by sex.

\n\n\nMaterials and methods\n\nSubjects, imaging acquisition, and ROI definition\n

Three + different datasets were considered. Table + 1 contains a comparative listing of the number of subjects per age + group in each dataset. The age groups have been selected on (approximately) + a decade basis. A subject division by sex is also provided and will be necessary + for the assessment of FC differences among males and females during aging. + The first dataset consists of participants from the International Neuroimaging + Data-sharing Initiative (INDI), 1000 Functional Connectomes\u2019 Project + (1000FCP) [13, + 14]. The cohort + contains N = 887 individuals from 17 centers with the requirement that a center + provide a minimum of 10 participants with full brain anatomical and functional + coverage. The ages of the subjects spanned 21\u201385 years, and 514 were + female. All subjects were at rest and not performing any task during recording. + Some subjects had their eyes open while some had their eyes closed, and each + recording consisted of signals collected for 42 ROIs. Our analysis is based + upon preprocessed data of ROIs obtained from a previous study. We provide + an overview of the most relevant section from the methods of [11]. Seed-based analysis was performed + as previously described [15, + 16] to identify + seven networks. The following seed regions were used to delineate the networks: + left posterior cingulate cortex, right frontoinsula, right intraparietal sulcus, + right superior parietal cortex, left auditory cortex, right visual cortex, + and left motor cortex. Each seed was defined as a 6 mm radius sphere centered + on previously published foci. Correlation maps were produced by extracting + the time course from each of the above seeds. Then, the Pearson correlation + coefficient (PCC) between the time course and the time course of each voxel + across the whole brain was computed to create the voxel-wise connectivity + maps. The maps from a young age group (N = 458, 21\u201330 years) were used + to identify peak coordinates of additional ROIs representing each network. + All regions were defined as 6 mm radius spheres around the peak coordinate. + Information regarding the participants, number of time points collected for + each recording, and the rsfMRI acquisition and epidemiological parameters + of the different centers are detailed in Table + 2 as well as Tables 1 + and 2 of [11].

\n\n10.1371/journal.pone.0300720.t001\n An itemization of the number of subjects per age + group associated with rsfMRI data via the 1000FCP (N = 887), NKI-RS recording + center of 1000FCP (N = 307), SRPBS (N = 709), and camCAN (N = 652) datasets. +

The number of subjects per age group are bifurcated by sex via the convention + (male, female). *The 1000FCP subjects in Group #1 were, more precisely, in + the 21\u201330 range.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Group + #Age group (years)1000FCP,# + of subjectsSRPBS,# of subjectscamCAN,# of subjectsNKI-RS + (1000FCP),# of subjects
118\u201330458 + (215, 243)*327 (227, 100)79 + (35, 44)65 (37, 28)*
231\u20134085 (44, 41)126 (81, 45)105 (56, 49)27 (8, 19)
341\u201350119 (40, 79)115 (48, 67)101 (43, 58)71 (16, 55)
451\u201360119 (38, 81)69 (29, 40)101 (54, 47)67 (12, 55)
561\u20137067 (22, 45)61 (30, 31)104 (56, 48)45 (13, 32)
671\u201380--117 (55, 62)-
671+39 + (14, 25)11 (8, 3)-32 (11, 21)
781+--45 (23, 22)-
\n\n\n\n10.1371/journal.pone.0300720.t002\n (Top) Recording center demographics for the N = + 887 subjects from the 1000FCP that were used in the analysis.

Additional + details such as the number of slices, voxel size, and subject handedness can + be found in Table 1 + of [11]. (Middle) + Recording center demographics for the N = 709 subjects from SRPBS, with additional + information available in Table 5 of [17]. + (Bottom) Information on the N = 652 subjects from the camCAN dataset that + consisted of a single recording center. Further details about the subjects + and study can be found in [18, + 19].

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Recording + centerSubjects (male, female)Age + (years)Scanner, TR (s)Time + pointsEyes
NKI-RS307 (97, 210)21\u2013853T, 2.5120Open
Beijing119 (48, 71)21\u2013263T, 2225Closed
Cambridge101 (40, 61)21\u2013303T, 3119Open
COBRE66 (47, 19)21\u2013653T, 2150NA
Milwaukee43 (14, 29)44\u2013653T, 2175NA
New York32 (18, 14)22\u2013493T, 2192Open
St. Louis31 (14, 17)21\u2013293T, 2.5127Open
Atlanta28 (13, 15)22\u2013573T, 2205Open
Berlin26 (13, 13)23\u2013443T, 2.3195Open
Cleveland26 (9, 17)24\u2013603T, 2.8127Closed
Dallas21 (10, 11)21\u2013713T, 2115NA
Queensland18 (11, 7)21\u2013343T, 2.1190Open
Orangeburg17 (13, 4)25\u2013551.5T, 2165Closed
Palo Alto17 (2, 15)22\u2013463T, 2245NA
Munich14 (9, 5)63\u2013731.5T, 372Closed
Leiden 110 (10, 0)21\u2013273T, 2.18215Closed
Leiden 211 (5, 6)21\u2013283T, 2.2215Closed
Recording centerSubjects (male, + female)Age (years)Scanner, + TR (s)Time pointsEyes
Kyoto234 (141, 93)18\u2013783T, 2.5240Open
ATR108 (88, 20)20\u2013303T, 2.5240Open
Osaka29 (21, 8)29\u2013733T, 2.5240Open
SWA101 (86, 15)19\u2013553T, 2.5244Open
Hiroshima237 (87, 150)20\u2013793T, 2143Open
Recording centerSubjects (male, female)Age (years)Scanner, TR (s)Time pointsEyes
camCAN652 (322, 330)18\u2013883T, 1.97261Closed
\n\n\n

The + rsfMRI data from N = 709 subjects (286 female) spanning ages 18\u201379 years + in the Japanese Strategic Research Program for the Promotion of Brain Science + (SRPBS) project [17] + was also used. Healthy control (HC) data was taken from five sites in Japan: + Osaka, Showa, Kyoto, and Hiroshima Universities as well as the Advanced Telecommunications + Research Institute (ATR). Each recording consisted of signals collected at + 140 ROIs of a HC. The voxel sizes were generally not the same across the five + sites, however, the values were similar by being between 3 mm x 3 mm x 3 mm + and 4 mm x 4 mm x 4 mm (please see Table 5 of [17]). + A scanning time of 10 minutes was used, and the subjects had their eyes open + and fixated on a point during the recording. The 9730 unique PCC values were + provided as part of the dataset, thus no additional processing was required + to attain the features. The acquisition and imaging parameters for SRPBS are + shown in Table 2 + and in more detail via Table 5 and 9 of [17].

\n

Lastly, + we used rsfMRI data that was collected at one recording center as part of + the Cambridge Centre for Aging and Neuroscience (CamCAN) study [18, 19]. + A single testing center provided N = 652 subjects (330 female) spanning ages + 18\u201388 years. The subjects had their eyes closed during the recordings + and followed the instruction to not think of any one thing in particular. + A voxel size of 3 mm x 3 mm x 4.44 mm was used with an acquisition time of + 8 minutes and 40 seconds. The acquisition and imaging parameters are further + detailed in Table 2 + of [19]. To compute + the FC from ROI recordings, the Craddock unsmooth parcellation was used to + attain 840 ROIs. The mean intensity across voxels was used. Although the parcellation + provided 840 ROIs, we considered 808 across each subject as 32 had missing + values for some of the subjects. The PCC values were computed by correlating + the time courses of the ROIs.

\n\n\nPreprocessing + and head motion correction\n

With the 1000FCP dataset, the functional + images were preprocessed in [11] + using the FSL software (FMRIB Software Library v. 5.0.1, Oxford, UK) and SPM + software (Statistical Parametric Mapping software package, Wellcome Department + of Imaging Neuroscience, London, UK) following conventional methods originally + described in [16, + 20]. The following + steps were preformed: 1) slice-timing correction; 2) rigid-body motion correction; + 3) registration to the MNI152 space; 4) regression of nuisance variables including + ventricles, white matter (WM), and global average signal 5) temporal bandpass + filtering (0.01\u20130.08 Hz). The preprocessing included rigid body correction + for motion within and across runs, normalization to the standard echo-planar + imaging (EPI) template of the Montreal Neurological Institute (MNI), and compensation + for slice-dependent time shifts. Several key acquisition parameters are shown + in Table 2 with + additional information such as the number of slices, voxel size, and subject + handedness available in Table + 1 of [11]. + The preprocessed functional data (in atlas space) were temporally filtered + to remove constant offsets and linear trends over each run while retaining + frequencies below 0.08 Hz. Data were spatially smoothed using a 4 mm full-width + half-maximum (FWHM) Gaussian blur. Sources of spurious or regionally non-specific + variance were removed by regression of nuisance variables. This included six + parameters obtained by rigid body head motion correction, the signal averaged + over the whole brain (global signal), the signal averaged over the lateral + ventricles, and the signal averaged over a region centered in the deep cerebral + WM. Temporally shifted versions of these waveforms were removed by inclusion + of the first temporal derivatives (computed by backward differences) in the + linear model. The analysis used via [11] + to account for head motion is summarized as follows. The parameters were calculated + for each participant as proposed by [22]. + Framewise displacement (FD), which represents head displacement from volume + to volume, was computed as the sum of the first derivative of the six rigid-body + motion parameters estimated during standard volume realignment. Delta variation + signal (DVARS), which represents the change in BOLD signal intensity from + one frame to the next, was computed as the root mean square average of the + first derivative of fMRI signals across the entire brain. A standardized version + of DVARS was applied according to [21]. + Volumes with FD value over 0.5\u2009mm or DVARS value over 1.5 IQRs above + the 75th percentile were removed with one prior and two subsequent volumes. + Sixteen individuals from several centers were excluded due to excessive head + motion during rsfMRI acquisition according to the described movement parameters + criteria [22].

\n

The + preprocessing of the SRPBS rsfMRI data are detailed in [17]. This was primarily done using SPM8 + implemented in MATLAB (R2016b; Mathworks, Natick, MA). The first 10 seconds + of data were discarded to allow for T1 equilibration. Preprocessing steps + included slice-timing correction, realignment, co-registration, segmentation + of T1-weighted structural images, normalization to MNI space, and spatial + smoothing with an isotropic Gaussian kernel of 6 mm FWHM. The ROIs were delineated + according to 140 regions defined by the Brainvisa Sulci Atlas and three subregions + of the cerebellum (the left and right cerebellum, and the vermis). The BOLD + signal time courses were extracted from the ROIs, and a bandpass filter (0.008\u20130.1 + Hz) was applied to the time courses before a regression procedure. Filtered + time courses were linearly regressed on temporal fluctuations of the WM, the + cerebrospinal fluid (CSF), and the entire brain, as well as six head motion + parameters. Fluctuation in each tissue class was determined from the average + time course of the voxels within a mask created by segmentation of the T1 + image. Extracted time courses were then bandpass filtered (0.008\u20130.1 + Hz) before linear regression, as was done for regional time courses. Then, + for each individual, a matrix of 9,730 connections between the 140 ROIs was + calculated by evaluating PCCs of BOLD signal time courses. The flagged frames + in the previous procedure were discarded. After calculating the FD, volumes + with FD > 0.5 mm were removed. Further details and information about the + procedure to calculate the matrix are outlined in [17, 21].

\n

The + camCAN preprocessing of the rsfMRI data was performed and is detailed in [19]. The procedure is + summarized as follows. Co-registered T1 and T2 images were used in a multi-channel + segmentation (SPM12 Segment, based on \u201CNew Segment\u201D in SPM8 [23]) routine in order + to extract probabilistic maps of six tissue classes: gray matter (GM), WM, + CSF, bone, soft tissue, and residual noise. Images from each subject were + coregistered to the subject\u2019s T1-weighted image using a rigid-body (6-df) + linear transformation. Normalization parameters from the diffeomorphic anatomical + registration through exponentiated lie algebra (DARTEL) procedure [24] were applied. The native-space GM and + WM images for all participants who passed quality-control checks were submitted + to diffeomorphic registration to create group template images. The group template + was then normalized to the MNI template via an affine transformation. Then + combined normalization parameters were applied to each individual participant\u2019s + GM and WM images. Individual normalized GM and WM images were smoothed (8 + mm FWHM Gaussian kernel). Data from each recording was unwarped to compensate + for magnetic field inhomogeneities, then realigned to correct for motion, + and slice-time corrected. The EPI data were co-registered to the T1 image, + and the normalization parameters were applied to warp functional images into + MNI space. The mean regional time-courses were then extracted using the template + method.

\n\n\nComputed features and description + of ML pipeline\n

To measure FC, PCC values were evaluated between + the time courses for all pairs of ROIs of a participant. Each recording consisted + of signals collected for 42, 140, and 808 ROIs for subjects in the 1000FCP, + SRPBS, and camCAN datasets, respectively. A simple calculation shows that + we have 861, 9730, and 326,028 unique PCC values to be used as features. The + features from each participant were used either as training or test data in + the ML analysis. We form two sets A and B, consisting of a younger age group + and an older age group for the scenarios considered in Fig 1. Let N1 denote the number + of subjects in the young group and N2 the number in the aged group. + The following steps are taken.

\n

Step 1) Let + m = min{N1, N2}.

\n

Step + 2) Train a machine on the m data points from the smaller set, + and m randomly selected points from the larger set. This is done to ensure + balance of data for an age range. The feature vectors will consist of the + PCC values for a subject. The labels correspond to whether the subject is + from the aged or young group.

\n

Step 3) Use the + k = max{N1, N2}\u2212m + left-out recordings from the larger set as validation data. The trained machine + is to predict whether the k residual recordings are young or aged and an accuracy + rate (AR) will be attained.

\n

Step 4) Test on + the subjects not included in sets A and B, but rather in interval i: i = 1, + 2, or 3 for the Fig 1 + scenario that is being considered.

\n\n10.1371/journal.pone.0300720.g001\n\n\nThree scenarios for assessing whether brain aging + is a progressive phenomenon.\n

In each case a machine is trained + on a young and aged cohort and asked to classify CPs of subjects with ages + that it has not previously encountered. For each scenario an illustration + contains possible hypotheses (H) for the trajectory of the predicted brain + ages in the respective interval. A) In interval 1 it is expected that the + percentage of brains classified as aged follow a monotonically increasing + trajectory among the chronological age range with the y-intercept being less + than 50%. B) The study of interval 2 investigates whether the CP of young + brains provide information about the CP of aged brains. C) Interval 3 is intended + to investigate whether CP differences among aged brains can be used to infer + aging dynamics of younger brains.

\n\n\n\n

The above steps are depicted in the pipeline + of Fig 2 where the + schematic is iterated once with the performance on the k residue data points + from the larger set viewed as validation accuracy. The trained machine is + then presented with the left-out data from the new category. The ML technique + used was a support vector machine (SVM) with a linear kernel and a binary + classifier. The techniques were implemented in R via the packages e1071 and + kernlab.

\n\n10.1371/journal.pone.0300720.g002\n\n\nAn ML pipeline to evaluate whether brains age + in a monotonic fashion.\n

The sets A and B refer to the young and + aged groups of the different scenarios that are considered when testing on + subjects of interval 1, 2, and 3. An SVM is trained on 2m data points, asked + to classify k residue data points from either the aged or young group, and + classify the age phenotype of the subjects in interval 1, 2, or 3.

\n\n\n\n\n\nMethodology + to assess monotonicity of brain differences with age\n

It is natural + to assume that a brain will possess characteristics that are more similar + to brains close to its age than those of increasingly disparate age groups. + A test of this notion is to present a trained machine with recordings across + a spectrum of ages that it has not previously seen. The three scenarios in + Fig 1 are considered + with the ML pipeline in Fig + 2. This will encompass training a machine on connectivity patterns + (CPs) from an equal number of subjects from a young and an aged cohort in + each scenario. The trained machines\u2019 decisions will scrutinize whether + brains\u2013whose age the machine has not previously encountered\u2013have + a CP that is more similar to an older or a younger cohort. The cartoons accompanying + each of the scenarios in Fig + 1 denote possible trajectories that we hypothesize for the left-out + recordings that will be classified by the respective machine. Interval 1 is + the most conspicuous since it is expected that as chronological age increases, + brains exhibit properties that become increasingly similar to an extrema group + of aged brains in a monotonic fashion. It is also expected that the youngest + brains in the interval exhibit CPs that are close to the extrema of young + subjects (i.e., an intercept < 50%). The described relation is referred + to as hypothesis 1 (H1) in Fig + 1A.

\n

The prospective hypotheses are less conspicuous for interval + 2. It is conceivable that older subjects exhibit CPs that follow an increasing + trajectory away from younger brains that have been bifurcated into a younger + (i.e., aged) and youngest (i.e., young) groups. This possibility is reflected + by the two increasing lines labeled H1 and H2 in Fig + 1B. The difference in the intercept of the two lines reflects the notion + that the youngest brains in interval 2 can be classified as being closer in + CP to brains in the aged group that are the most proximal to them in years + (H1), or alternatively as relatively young (H2) if the CPs in the unseen data + are classified on a relative rather than absolute basis. Lastly, it is conceivable + that the recordings exhibit CPs that do not deviate very much from the aged + (yet relatively younger) brains in which case a horizontal line with an intercept + greater than 50% would be noted (H3). The investigation of interval 2 probes + at whether CPs of young brains provide information on the dynamics noted in + older subjects.

\n

Interval 3 in Fig + 1C represents a somewhat inverted version of the study for interval + 2. It investigates whether older brains provide insight into the connectivity + changes that occur among younger brains. A hypothesis shown by the strictly + increasing line (H1) entails young brains exhibiting properties reminiscent + of the youngest brains in an otherwise aged cohort, and the CP of the young + brains gradually resembling that of the most aged brains as chronological + age is increased. It is also conceivable that the gradual resemblance to the + most aged brains does not occur, but rather the trained machine associates + the brains in interval 3 as resembling the young cohort as the chronological + age of the subjects approaches that of the subjects in the more proximal young + cohort. This scenario is depicted via the inverted U-shaped trajectory (H2). + Lastly, it may be that the CPs of subjects remain similar to the younger group + as shown by the trajectory of a horizontal line with an intercept below 50% + (H3). The investigation of interval 3 probes at whether aged brains provide + information on the dynamics of connectivity earlier in life.

\n\n\nQuantification of FC changes with age and sex\n

The + pairwise PCC values were used as the features for the ML analysis and to assess + the connectivity among brain regions. For each subject the number of connections + that exceeded the threshold |\u03C1 = 0.6| were totaled and referred to as + the FC for the subject. The value of 0.6 has been motivated in works such + as [19, 25]. For completeness, thresholds of |\u03C1 + = 0.45| and |\u03C1 = 0.7| are also considered in the analysis to assess the + robustness of the findings. The evaluation with different FC thresholds is + to account for the possibility of low SNR in the recordings. Typically, signals + in a stable or high SNR regime would provide similar trends when a correlation + threshold is uniformly adjusted. Importantly, we consider the positive and + negative (i.e. anticorrelation) CC values separately when assessing the change + in connectivity with aging for the three datasets. In testing for the statistical + significance of an effect, the youngest age group is taken as a reference + and a two-sample t-test is conducted among the positive, negative, and total + connections in the subjects. A p-value of 0.05 is used as the threshold for + statistical significance when evaluating the change in FC between the youngest + age group and the older subjects. The change is assessed separately among + male and female subjects with the objective of determining whether age-modulated + FC is sex specific. As done with the entire population, the youngest age group + is considered the reference for assessing FC changes, and a two-sample t-test + is used to test for statistical significance (i.e. p < 0.05). The standard + deviations (s.d.) of the number of functional connections are also computed + to study the inter-subject variability in connectivity during aging.

\n\n\nConsideration of recording center and subject arousal + effects\n

The 1000FCP dataset consists of recordings from 17 centers, + and thus warrants scrutiny of potential site effects. The NKI-Rockland (NKI-RS) + center contains approximately 34% of the subjects in the 1000FCP. This number + is significant when considering the next largest participant pool (i.e. Beijing) + contributes approximately 13%. NKI-RS also contains the largest spread of + subjects\u2019 chronological ages (Table + 1). The monotonicity and change in FC analysis shall be repeated with + only subjects from the NKI-RS center. A comparison of the findings for 1000FCP + and its constituent NKI-RS will show whether the same trends are noted for + both cases or if site effects may be affecting the findings. Furthermore, + unlike SRPBS and camCAN, the 1000FCP contains a mixture of recordings where + subjects had their eyes open and closed (Table + 2). A difference in arousal level can affect FC strength and bias connectivity + patterns. The monotonicity and change in FC analyses are conducted separately + on 1000FCP recordings where subjects had their eyes open (N = 543). The reason + for not additionally conducting an analysis on subjects with eyes closed is + because of their relatively low number (N = 197), and lack of representation + across the spectrum of ages in the 1000FCP dataset (Table 2).

\n\n\n\nResults\n\nMonotonic + changes in functional connectivity occur over an intermediate age interval\n

In + considering the 1000FCP data, the scenario in Fig + 1A encompassed a machine trained on features from 78 brains, 39 that + comprise the 71+ age group and 39 of which have been randomly selected from + the 21\u201330 group. The remaining 809 brains shall be considered as test + data. Naturally, it is expected that the 419 remaining recordings from the + 21\u201330 age group be classified as young. In evaluating the accuracy on + the test data, excellent performance is seen for the left-out young brains + in the 21\u201330 age group via an AR of 0.84 across 419 subjects. The accuracy + vindicated the machine\u2019s integrity to continue the monotonicity study + and evaluate decisions on recordings in interval 1. The decisions made by + the machine on the 390 brains of ages 31\u201370 years are shown in Fig 3. We have considered + the number of brains at each age that were classified as old to arrive at + percentages for the ages encompassed in interval 1. The bar plot illustrates + the number of brains at an age that have been classified as young or aged. + The accompanying line plot represents the percentage of subjects that were + classified as aged after the application of a 5-year sliding window for smoothing. + Although the increase is not monotonic, an increasing trend is noted in the + percentage. The absence of complete monotonicity is due to cross-subject variability + as variance inherently exists among all facets of recordings from different + people. Furthermore, there is the limitation that the number of samples at + each age in Fig 3 + was not very large (min sample size = 3, max sample size = 19). Such confounding + factors are natural and expected to provide uncertainty to the findings. This + is noted by the moderate Spearman correlation value of 0.7 between the subjects\u2019 + ages and the percentage of subjects classified as aged. To test for a trend, + a linear regression with intercept \u03B20 and slope \u03B21 + was fit to the results. As previously mentioned, a positive trend and a low + intercept are expected for the brains in interval 1. The intercept \u03B20 + = 31.58 is rather low indicating that the classification of the youngest brains + in interval 1 are indeed skewed towards the young group. A slope of \u03B21 + = 0.952 demonstrates a positive association between the increase in age and + CPs resembling the older phenotype. The results justify H1 as holding for + interval 1 of Fig 1A.

\n\n10.1371/journal.pone.0300720.g003\n\n\nThe results of the brain age monotonicity analysis + for interval 1.\n

The histograms show the number of subjects classified + as young and aged at every considered age. A 5-year sliding window was applied + and a linear regression was fit to the results with intercept \u03B20 + and slope \u03B21. A machine was trained on the youngest and oldest + groups and presented with test data of intermediate-aged subjects. For 1000FCP + we have \u03B20 = 31.58, \u03B21 = 0.952. The Spearman + correlation of 0.7 computed among the percentages of subjects classified as + aged and the ages indicates a moderate degree of monotonicity between the + two quantities. With SRPBS, \u03B20 = 7.285, \u03B21 + = 2.696, and a Spearman correlation of 0.948 demonstrates a high degree of + monotonicity between the two quantities. A consideration of camCAN leads to + \u03B20 = 0 and \u03B21 = 2.234. The Spearman correlation + of 0.986 signifies monotonicity between the two quantities. For NKI-RS subjects + we note \u03B20 = 39.74, \u03B21 = 0.877, and the Spearman + correlation of 0.608 reflects monotonicity.

\n\n\n\n

For the SRPBS subjects, a machine is trained + on 72 subjects randomly selected from the 18\u201330 age group and the 72 + in the 61+ group. The test set consists of 565 subjects, 255 of which are + residual from the 18\u201330 year old (y.o.) range. The machine classified + 247 of the residual recordings as young leading to AR = 0.968 and vindicating + the machine efficacy. The histograms in Fig + 3 illustrate the number of subjects assigned as young and aged at every + considered age. To test for a trend, a linear regression was applied to the + results in interval 1. The low intercept (\u03B20 = 7.285) and + the positive slope (\u03B21 = 2.696) were expected while a Spearman + correlation of 0.948 indicates a high degree of monotonicity in the brains + classified as having an aged profile with increasing biological age. The hypothesis + H1 in Fig 1A is validated. + Similarly, for the camCAN dataset a machine was trained on 45 randomly selected + subjects from the 18\u201330 age group and the 45 subjects 81+ y.o. Of the + 34 remaining subjects in the younger group 33 were classified as young leading + to AR = 0.9705. A linear fit to the percentage of subjects classified as aged + in Fig 3 yields the + intercept \u03B20 = 0 and slope \u03B21 = 2.234. The + Spearman correlation of 0.986 computed among the percentages of subjects classified + as aged and their chronological ages indicates monotonicity between the two + quantities. The analysis supports H1 as the brain aging dynamics of intermediate + aged subjects when extrema groups are used as the references. The NKI-RS analysis + entailed a machine trained on features from 32 recordings of subjects in the + 71+ age group and 32 randomly selected from the 21\u201330 age group. The + remaining 33 recordings from the young group (21\u201330 y.o.) were classified + by the trained machine at an AR of 0.696. The decisions on the residual 210 + brains spanning 31\u201370 years are shown in Fig + 3. An increasing trend (\u03B21 = 0.877), relatively low + intercept (\u03B20 = 39.74), and a formidable Spearman correlation + value of 0.608 are noted. The results also support hypothesis H1 as an explanation + for brain aging dynamics in interval 1.

\n\n\nYoung + brain functional connectivity is not indicative of alterations in older subjects\n

Considering + the 1000FCP subjects, the ML pipeline is applied with the partition in Fig 1B to probe whether + there is a monotonic relationship. A machine is trained on 170 brains, 85 + of which have been randomly selected from the 21\u201330 age group, and 85 + that comprise the 31\u201340 age group. The remaining 717 subjects in the + dataset are used to investigate whether brains are perceived as young or aged + by the machine. The 373 remaining brains from the 21\u201330 age group are + expected to be classified as young, however, the decisions are not obvious + for the 344 brains that constitute interval 2. The AR noted is 0.646 for the + held-out brains in the young group. It is expected that the AR levels for + Fig 1B be lower than + those reported for Fig 1A\u2013this + is attributed to the difficulty in distinguishing among brains that are closer + in age. We consider the decisions made by the trained machine on the 344 elder + brains (41\u201371+ y.o.) of interval 2. In Fig + 4 the percentage of brains classified as aged does not show a monotonically + increasing relationship (Spearman correlation = -0.243). The point 77+ on + the abscissa is considered as the final age because of the relatively few + samples available above this age\u2013the point is comprised of 14 subjects + ranging from 77 to 85 y.o. When fit with a linear regression, the intercept + \u03B20 = 71.4 is rather high indicating that the youngest brains + in interval 2 appear more similar to the proximal brains in the aged group + rather than the relatively young group that was used during training. The + slope of \u03B21 = -0.147 does not support an increasing trend + for the 41+ y.o. subjects of interval 2. The Spearman correlation of -0.243 + between chronological age and the percentage of brains classified as aged + indicates that the change in CP dynamics of the younger brains is not conserved + among the progressively older brains. We conclude that hypothesis H3 in Fig 1B is the most representative + trajectory for interval 2.

\n\n10.1371/journal.pone.0300720.g004\n\n\nThe results of the brain age monotonicity analysis + for interval 2.\n

The histograms show the number of subjects classified + as young and aged at every considered age. A 5-year sliding window was applied + and a linear regression was fit to the results with intercept \u03B20 + and slope \u03B21. A machine was trained on the two youngest groups + of subjects and then asked to classify older recordings. For 1000FCP we attain + \u03B20 = 71.4 and \u03B21 = -0.147 indicating little + change in the classification of increasingly older brains as aged. Furthermore, + a Spearman correlation of -0.243 reflects a non-monotonic relationship. With + the SRPBS subjects \u03B20 = 82.159 and \u03B21 = 0.1 + suggest little change in the classification of increasingly older brains as + aged. The Spearman correlation of 0.149 reflects a non-monotonic relationship + among the ages and the percentage classified as aged. For camCAN we attain + \u03B20 = 77.48, \u03B21 = 0.329 demonstrating little + change in the classification of increasingly older brains as aged. The Spearman + correlation of 0.74 reflects a moderately monotonic relationship among the + subjects\u2019 ages and the percentage classified as aged. The consideration + of NKI-RS subjects of interval 2 leads to \u03B20 = 65.48, \u03B21 + = 0.453, and a Spearman correlation of 0.417. This indicates an increasing + relationship but little monotonicity.

\n\n\n\n

Applying the analysis to the SRPBS dataset + involved a machine trained on 126 subjects randomly selected from the 18\u201330 + age group and 126 comprising the 31\u201340 group. The test set consisted + of the remaining 457 subjects. Of the 201 residual subjects from the 18\u201330 + y.o. group, 146 were classified as young leading to AR = 0.726. In interval + 2 we attain intercept and slope values of \u03B20 = 82.159, \u03B21 + = 0.1 indicating little change in the appearance of increasingly older brains + from the more proximal aged group. A Spearman correlation of 0.149 in Fig 4 denotes negligible + monotonicity in the percentage of subjects classified as aged with an increasing + number of years. The findings support hypothesis H3 that studying the aging + trajectory in younger brains (i.e. 18\u201340 y.o.) provides little information + about their aging trajectories in later years. Analysis with the camCAN subjects + considered the training data comprised of 79 subjects from the 18\u201330 + as well as the 31\u201340 age groups, and testing done on the 493 residual + subjects. The 26 subjects from the older group were classified as aged at + an AR of 0.653 by the trained machine. After computing the percentages of + 468 subjects deemed as old at each age in the 41+ range, the linear fit provided + \u03B20 = 77.48 and \u03B21 = 0.329. This indicates + little increase in the classification of the older subjects with the increase + in their chronological ages. The Spearman correlation of 0.74 between the + percentage of brains classified as aged and the chronological age reflects + a moderately monotonic relationship. The findings signify H1 as the most expressive + of the hypotheses for the trajectory in interval 2. NKI-RS subjects in the + age group 21\u201330 and 31\u201340 were used to train an SVM for the interval + 2 analysis. Following the selection of 27 random recordings from the younger + group, the 38 left out recordings were classified at a below-chance level + of AR = 0.473. By computing the percentages of the 215 subjects that were + 41+ y.o. and classified as aged, a linear fit provided \u03B20 + = 65.48 and \u03B21 = 0.453 (Fig + 4). While the slope indicates an increasing trend, a Spearman correlation + of 0.417 between the percentage of brains classified as aged and the subjects\u2019 + ages in interval 2 reflects the absence of a monotonic relationship. The analysis + signifies H3 as the most representative hypothesis for the NKI-RS trajectory + in interval 2.

\n\n\nConnectivity patterns + of older subjects reflects aging dynamics in young brains\n

Through + the scenario depicted in Fig + 1C we explore if older brains exhibit CPs that can be used to differentiate + among younger brains, and whether the differentiation occurs in a monotonic + fashion with age. We hypothesize the trained machine to provide a low intercept + when investigating a trend in the percentage of brains classified as aged. + However, it is not obvious if a monotonic increase or a positive association + will be seen. For the 1000FCP dataset an SVM is trained on 78 brains; 39 randomly + selected from the 61\u201370 age group and 39 from the 71+ age group. The + residual subjects from the 61\u201370 age group are expected to be classified + as young, and an AR of 0.714 was noted on the 28 held-out brains from this + group. The decisions made by the trained machine on the 781 younger brains + (ages 21\u201360 years) are shown in Fig + 5. The intercept of \u03B20 = 24.2 is somewhat high since + the ideal intercept would be near zero. This demonstrates that the relatively + young brains that were used to train the machine are not very representative + of the FC changes in the youngest subjects of the study. The slope in Fig 5 is \u03B21 + = 0.58 indicating a marginal but increasing trend in the percentage of brains + classified as aged across interval 3. A Spearman correlation of 0.795 is noted + among the subjects\u2019 ages and the percentage of brains classified as aged. + We conclude that H1 is the best hypothesis for interval 3. This indicates + that the difference in CP among aged brains is reflective of FC dynamics earlier + in life.

\n\n10.1371/journal.pone.0300720.g005\n\n\nThe results of the brain age monotonicity analysis + for interval 3.\n

The histograms show the number of subjects classified + as young and aged at every considered age. A 5-year sliding window was applied + and a linear regression was fit to the results with intercept \u03B20 + and slope \u03B21. A machine was trained on the two oldest age + groups and received younger subject recordings as test data. With the 1000FCP + dataset, the values \u03B20 = 24.2, \u03B21 = 0.58, + and the Spearman correlation of 0.795 indicate a gradual increase and moderate + monotonicity. For SRPBS, \u03B20 = 3.11, \u03B21 = 0.788, + and the Spearman correlation of 0.853 demonstrates monotonicity in the number + of subjects classified as aged with increasing age. With camCAN, the values + \u03B20 = 32.52, \u03B21 = -0.205, and the Spearman + correlation of -0.448 indicate a gradual decrease and small degree of monotonicity. + The NKI-RS subjects of interval 3 provide \u03B20 = 45.12 and \u03B21 + = -0.176 to reflect a gradually decreasing relationship. The Spearman correlation + of -0.161 indicates an absence of monotonicity.

\n\n\n\n

Applying the ML pipeline to the SRPBS + dataset, a machine is trained on 69 subjects in the 51\u201360 age group and + 69 subjects randomly selected from the 61+ group. The testing was performed + on the remaining 571 subjects. Of the 3 remaining subjects from the 61+ y.o. + group, 2 were correctly classified as aged (AR = 0.667). By applying the ML + pipeline of Fig 2 + to this scenario, we attain the results in Fig + 5 for interval 3. A linear fit provides a low intercept (\u03B20 + = 3.11) and a positive slope of \u03B21 = 0.788. The Spearman correlation + of 0.853 signifies that differences among aged brains (i.e. 61+ y.o.) provide + a high degree of information about the aging trajectory of younger brains. + The quantified results indicate that hypothesis H1 is the most explanatory + of the three hypotheses for interval 3. The study with the camCAN subjects + involved training on 45 young (71\u201380 y.o.) and 45 aged (81+ y.o.) subjects + and subsequently testing on the 561 residual recordings. The 72 residue subjects + in the 71\u201380 group were classified accurately as young via AR = 0.666. + For the remaining 489 recordings that comprise interval 3, the linear fit + provides \u03B20 = 32.52 and \u03B21 = -0.205. The Spearman + correlation of -0.448 reflects a small degree of monotonicity in the number + of subjects classified as aged when the reference was taken to be the aging + dynamics of older brains. The scenario H3 is the endorsed hypothesis as the + properties of the two chronologically oldest groups do not markedly differentiate + between the ages of the recordings from the younger subjects. The ML analysis + for the NKI-RS subjects involved training on 32 young (61\u201370 y.o.) and + 32 aged (71+ y.o.) recordings. The 13 residual subjects in the 61\u201370 + group were classified as young via AR = 0.692. The decisions made by the machine + on the 230 younger subjects comprising interval 3 yield \u03B20 + = 45.12 and \u03B21 = -0.176 when fit with a linear model (Fig 5). A Spearman correlation + of -0.161 does not reflect monotonicity and H3 is recognized as the most supported + hypothesis.

\n\n\nFunctional connectivity + changes are categorical with aging\n

In light of the ML analysis + providing predictions from measures of FC, we pursue a quantitative evaluation + of the FC and how it varies across the age spectrum. For the 1000FCP dataset, + Fig 6A shows the mean + and standard deviation (s.d.) of the FC for subjects across the seven decades. + A significant decrease in mean FC in noted among the 51\u201360, 71\u201380, + and 81+ groups (p < 0.05). The positive FC values that showed a significant + decrease coincided with the total FC except for the 31\u201340 age group. + Although less apparent, the presence of anticorrelations was not negligible, + and the decrease in the negative FC values was significant for the 81+ group. + The alteration in FC with aging was also prevalent for the SRPBS subjects, + however the direction of the change was different as there is an increase + in FC with age (Fig 6B). + The increase accelerated among the 51\u201360 and 71+ groups and, aside for + the positive correlations in the 61\u201370 y.o. group, is seen for positive, + negative, and total connectivity measures. The change in the number of negative + connections is not insignificant and follows an increasing trend that coincides + with the conventionally considered (i.e. positive) correlations. The camCAN + subjects show a decrease in FC that is steady but significant with increasing + age (Fig 6C). Interestingly, + while being negligible in value in comparison to the number of positive correlations, + the anticorrelations show an opposite trajectory by increasing with aging. + For the NKI-RS dataset, Fig + 6D shows the mean of the FC for subjects. A significant decrease in + mean FC is noted among the 81+ y.o. group and for the positive FC values of + the 71\u201380 y.o. group (p < 0.05).

\n\n10.1371/journal.pone.0300720.g006\n\n\n\n<p>A study of the change in FC for subjects + across decades from the A) 1000FCP, B) SRPBS, C) camCAN, and D) the NKI-RS + recording center. The mean FC was computed by averaging across subjects the + number of times that the Pearson CCs from the connectivity matrix exceeded + a threshold of |\u03C1| = 0.6. The bifurcation of the FC in positive and negative + (i.e. anticorrelations) directions is also shown. Two-sample t-tests are used + to assess the significance of the change in FC with the youngest age group + taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons between positive, negative, and total + FC values. The s.d. of the number of functional connections is illustrated + to study the inter-subject variability across the age spectrum.</p>\n</caption>\n<graphic + mimetype=\"image\" position=\"float\" xlink:href=\"info:doi/10.1371/journal.pone.0300720.g006\" + xlink:type=\"simple\"/>\n</fig>\n<p>We examine the inter-subject variability + in FC with aging. In the case of 1000FCP, there is an increase in variability + among middle age to elder (61\u201370) subjects, and a reduction in FC variability + for the two oldest decades (<xref ref-type=\"fig\" rid=\"pone.0300720.g006\">Fig + 6A</xref>). It is intriguing that the increase in FC consistency for the oldest + subjects is precipitous and present among the positive and negative correlations. + Similarly, the peak inter-subject FC variability occurs in middle age for + the SRPBS subjects (51\u201360 y.o.) with decreasing variance for older and + younger age groups (<xref ref-type=\"fig\" rid=\"pone.0300720.g006\">Fig 6B</xref>). + The youngest subjects (18\u201330 y.o.) exhibit the most consistency. The + variance in the inter-subject FC for the camCAN subjects shows a steady decrease + along the decades. There is a bimodal division as the older subjects (61+ + y.o.) have a more consistent CP than the younger subjects (<xref ref-type=\"fig\" + rid=\"pone.0300720.g006\">Fig 6C</xref>). It is interesting that the datasets + provide very different results on the variability in inter-subject FC across + the broad age spectrum. While the 1000FCP and camCAN subjects show the least + variability in the oldest age groups, the SRPBS subjects show the lowest variability + for the youngest subjects. Furthermore, the 1000FCP and SRPBS subjects show + the highest variability for middle-age (61\u201370 and 51\u201360, respectively), + while the youngest decade (18\u201330 y.o.) camCAN subjects exhibited the + largest inter-subject FC variability. In the case of the NKI-RS subjects there + is an increase in variability among middle age to elder (61\u201370 y.o.) + subjects, and a reduction in FC variability for the two oldest decades (<xref + ref-type=\"fig\" rid=\"pone.0300720.g006\">Fig 6D</xref>). The trend is very + similar to the aggregate 1000FCP scenario since the increase in FC consistency + is precipitous for the oldest subjects and exists for positive and negative + correlations. In summary, the change in the total FC is categorical with aging + across the populations of subjects. The inter-subject variance, however, follows + an inverted U-shaped trajectory in two datasets, and a bimodal trajectory + showing a decrease in variability with aging in the third case.</p>\n</sec>\n<sec + id=\"sec014\">\n<title>Brain aging dynamics are similar among sexes, but accentuated + in males\n

In considering the 1000FCP subjects, the results in Fig 7 indicate that a decrease + in FC with increasing age is more pronounced in males. Specifically, the two + middle-age groups of males (41\u201350, 51\u201360) exhibited a significant + decrease in their positive FC measures (p < 0.05) while only the 51\u201360 + y.o. group of females showed a similar decline. The significant decrease in + FC is noted in the total FC of males for the two oldest groups. A similar + decrease is noted in females of 81 years and older. Interestingly, the oldest + females witness a greater reduction in FC than the male subjects. For the + SRPBS dataset there is an increase in FC with aging for both males and females, + although the increase is only significant for the male subjects. Specifically, + the increase in all three classes of correlations are significant for the + male subjects in the 41\u201350 and 51\u201360 y.o. range, while only the + negative correlations are significant for the female subjects of the same + age groups (Fig 8). + Perhaps more importantly, the males show their largest number of negative + and total connections in the oldest age group. The subjects in the camCAN + cohort exhibit a similar trend with the progressive change in FC being more + prominent in males. While a significant decrease in total FC is noted in the + 71\u201380 and 81+ age groups for both sexes, the change starts earlier in + males via the 61\u201370 y.o. group (Fig + 9). The earlier alteration to FC in males is also noted among their + anticorrelations across the age spectrum. The change in FC for the NKI-RS + subjects is similar to that seen for the aggregate 1000FCP dataset (Fig 10). The decrease in + connectivity is only significant for the oldest age group of male and female + subjects. The change is more pronounced in males because the total and positive + FC values are statistically significant, while only the positive FC values + are significant for females.

\n\n10.1371/journal.pone.0300720.g007\n\n\nAn assessment of the differences in FC between + male and female subjects in the three datasets.\n

The mean and s.d. + of the number of connections are plotted separately for the male and female + subjects from the 1000FCP dataset. The mean FC was computed by averaging across + subjects the number of times the Pearson CCs from the connectivity matrix + exceeded a threshold of |\u03C1| = 0.6. The division of the FC in positive + and negative (i.e. anticorrelation) directions is also shown. Two-sample t-tests + are used to assess the significance of the change in FC with the youngest + age group taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons between positive, negative, and total + FC values. The s.d. of the number of FCs is shown to study the inter-subject + variability for each sex and age group.

\n\n\n\n\n10.1371/journal.pone.0300720.g008\n\n\nAssessment of the differences in FC between male + and female subjects in the three datasets.\n

The mean and s.d. of + the number of connections are plotted separately for the male and female subjects + from the SRPBS dataset. The mean FC was computed by averaging across subjects + the number of times the Pearson CCs from the connectivity matrix exceeded + a threshold of |\u03C1| = 0.6. The division of the FC in positive and negative + (i.e. anticorrelation) directions is also shown. Two-sample t-tests are used + to assess the significance of the change in FC with the youngest age group + taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons between positive, negative, and total + FC values. The s.d. of the number of FCs is shown to study the inter-subject + variability for each sex and age group.

\n\n\n\n\n10.1371/journal.pone.0300720.g009\n\n\nAssessment of the differences in FC between male + and female subjects in the three datasets.\n

The mean and s.d. of + the number of connections are plotted separately for the male and female subjects + from camCAN. The mean FC was computed by averaging across subjects the number + of times the Pearson CCs from the connectivity matrix exceeded a threshold + of |\u03C1| = 0.6. The division of the FC in positive and negative (i.e. anticorrelation) + directions is also shown. Two-sample t-tests are used to assess the significance + of the change in FC with the youngest age group taken as the reference. A + p-value of 0.05 is the threshold for statistical significance in the pairwise + comparisons between positive, negative, and total FC values. The s.d. of the + number of FCs is shown to study the inter-subject variability for each sex + and age group.

\n\n\n\n\n10.1371/journal.pone.0300720.g010\n\n\nAssessment of the differences in FC between + male and female subjects in the three datasets.\n

The mean and s.d. + of the number of connections are plotted separately for the male and female + subjects from the NKI-RS recording center. The mean FC was computed by averaging + across subjects the number of times the Pearson CCs from the connectivity + matrix exceeded a threshold of |\u03C1| = 0.6. The division of the FC in positive + and negative (i.e. anticorrelation) directions is also shown. Two-sample t-tests + are used to assess the significance of the change in FC with the youngest + age group taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons between positive, negative, and total + FC values. The s.d. of the number of FCs is shown to study the inter-subject + variability for each sex and age group.

\n\n\n\n

We examine the inter-subject variability + in FC with aging separately among the sexes. A rising and precipitous fall + in variability is noted among both sexes for the 1000FCP subjects (Fig 7). However, the peak variability occurs + at the different junctures of 51\u201360 for males and 61\u201370 for females + when considering the total FC. A similar inverted U-shape trend is noted with + the SRPBS data (Fig 8) + but with the peak variance in inter-subject FC at the 61\u201370 age group + in males while the peak variability for females occurred during the 51\u201360 + interval. In Fig 9, + the camCAN results for the female subjects showed a similar spike in inter-subject + variability during middle age since the 41\u201350 and 51\u201360 year-old + females had the least consistency. The males show a variation in FC that is + incongruous to what we have noted thus far. Namely, they showed the greatest + variance in the young (18\u201330 and 31\u201340) and a relatively constant + variance in the latter decades. The NKI-RS male and female subjects show a + rising and precipitous falling in FC variability with aging (Fig 10). The trends are similar to that for + the aggregate 1000FCP since the peak variabilities occur at 51\u201360 for + males and 61\u201370 for females. In both sexes the lowest variability is + seen for the oldest age group.

\n\n\n\nDiscussion\n

In + the course of healthy human aging, one may expect a monotonic relationship + with young brains having properties that are increasingly similar to aged + brains. The application of ML pipelines can aid in the discovery of fundamental + facets of brain aging including the role of sex. While our use of SVM was + motivated by prior works [5, + 26], the questions + and classification-based methodology are novel. The three intervals studied + in Fig 1 investigate + questions pertaining to the dynamics of brain aging among healthy subjects. + A strong degree of monotonicity was noted when considering the two extrema + age groups as references (interval 1 of Fig + 1A). This was consistent among three datasets that contained subjects + of diverse demographics. A discrepancy from a Spearman correlation value of + one\u2013i.e., a completely monotonic relationship\u2013points towards several + practical considerations. For instance, the study is prone to inter-subject + variability since participants of the same age may have very different genetic + traits, environmental factors, and life histories while participants in a + different age group may be more similar to each other with respect to the + aforementioned factors.

\n

The analysis that accommodated a new question + of whether the CPs of young brains contain structure that is conserved among + progressively older brains (interval 2 of Fig + 1B) provides a less categorical answer. While the majority of brains + were classified as more similar to the aged group, there was little indication + of monotonicity in the classifiers\u2019 decisions across the age spectrum. + The difference in CP among younger brains was not indicative of nor resonated + to the CP changes of the older brains. Conversely, it is interesting that + information can be gained about the connectivity of young brains through the + CPs of strictly older brains (i.e. interval 3 of Fig + 1C). In two of the datasets a linear fit showed that the CP associated + with old subjects (61+ y.o.) are informative of relative changes in much younger + subjects. Table 3 + contains the prominent hypotheses among the possible brain aging trajectories + in each interval for the datasets. The ML parameters as well as the results + attained via the post-processing are also listed. To the best of our knowledge, + this work presents the first evaluation of monotonicity in brain aging and + its quantification via ML. The analysis has gone beyond the paradigm of brain + age prediction from fMRI signals or changes in brain connectivity to probe + whether differences in CPs of young brains are reflective of the differences + in later life. Interestingly, the complementary question of whether the difference + in CPs of aged brains are informative of the changes seen by younger subjects + provided a different answer. For each of the three intervals in the monotonicity + study, a 5-year sliding window was applied for smoothing. This was done prior + to fitting a linear regression to the percentage of subjects that were classified + as aged across the spectrum. The analysis was also performed without the sliding + window (S1 + Fig) and we did not witness a change in the conclusions. It is noteworthy + that the Spearman CC values were smaller in magnitude without the sliding + window. This is expected because not smoothing may obfuscate or hide trends + if the data at a single age contains outliers or when fewer, non-representative + subjects exist at an age. The interval 1 subjects in Figs 3\u20135 + that deviate from a monotonic relationship deserve further discussion. This + is explained by the limitation in the number of subjects as well as the empirical + nature of the collected data. Despite a smoothing operation having been applied, + the percentage of subjects classified as aged was computed for every year + in the interval. This temporal resolution further contributed to specific + ages consisting of a fewer number of subjects and the likelihood of a non-representative + CP affecting the results at a particular age. Another factor follows from + the choice of the monotonicity measure. Although a commonly used metric, the + Spearman CC is a measure of linear monotonicity. Conversely, the BOLD signal + as well as the aging process are nonlinear processes at both the micro- and + macroscale. As increased data, analysis, and results come forth, it will perhaps + be possible to thoroughly assess the suitability of nonlinear measures of + monotonicity for explaining brain aging dynamics.

\n\n10.1371/journal.pone.0300720.t003\n The comparative parameters and results of the monotonicity + study conducted with the three datasets.

The intercept (\u03B20) + and slope (\u03B21) attained via a linear fit to the percentage + of subjects classified as aged in intervals 1, 2, and 3 are included. The + Spearman CC and fitted parameters were used to determine the prominent hypothesis + among the possible brain aging trajectories.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
IntervalDatasetSize of training setSize of test set\u03B20\u03B21Spearman + CCProminent hypothesis
11000FCP7839031.580.9520.7H1
1SRPBS1443107.2852.6960.948H1
1camCAN9052802.2340.986H1
1NKI-RS6421039.740.8770.608H1
21000FCP17034471.4-0.147-0.243H3
2SRPBS25225682.1590.10.149H3
2camCAN15846877.480.3290.74H1
2NKI-RS5421565.480.4530.417H3
31000FCP7878124.20.580.795H1
3SRPBS1385683.110.7880.853H1
3camCAN9049032.52-0.205-0.448H3
3NKI-RS6423045.12-0.176-0.161H3
\n\n\n

While + quantitative ML analysis of how brain connectivity dynamics are modulated + by aging is important, it is also necessary to be cognizant of the features + that are input to the predictive algorithm. The Pearson CC has been used as + a measure of the degree and direction of the relationship\u2013i.e. connectivity\u2013among + pairs of ROIs. This measure and its transformations are prevalent in rsfMRI + works that investigate the effects of aging on CPs. Drawing upon several recent + studies, in [27] + task-evoked and rsfMRI were collected from subjects spanning 20 to 93 years. + The connectivity comparisons concluded that large-scale brain organization + was maintained during aging, but parcellation differences existed among age + groups and increased with larger differences in age. Conversely, the resting + and task fMRI data analyzed in [28] + led the authors to conclude that FC was lower in older adults. Correlational + analysis in [29] + also noted that aging led to a decrease in strong FC across the brain. The + decrease in modularity was quantified by more uniform activity patterns among + the ROIs of the older subjects. The more diffuse and fewer strongly correlated + functional connections with older age has been referred to as functional dedifferentiation + [30\u201332]. We also note a definitive + change in FC with increasing age that is more significant in the latter decades + of the subjects. Two of the datasets show the strength of the connections + to decrease (supporting functional dedifferentiation) while contrary results + were noted with the SRPBS cohort. The general decrease in the mean number + of significant CC values with increasing age is consistent with several prior + reports of the modularity in functional networks decreasing with aging [11, 29, 33, + 34]. A recent work + found mostly weakened connectivity with aging among subjects, but an increase + in several cases [8]. + The authors attribute the increases to the within network FC of several prominent + systems such as the visual and somatomotor. In [35] + a four-year longitudinal analysis of N = 16 elderly subjects (69+ y.o. at + the start of the study) was undertaken. The results revealed a decrease in + FC in two of the three considered brain networks. Such findings are interesting + in accommodating the non-unanimous results noted here when considering three + distinct datasets.

\n

The BOLD fMRI signal has been processed to quantify + FC in assessing a large scale of phenotypes studied in resting and task-based + studies. Surprisingly, much less attention has been given to its inter-subject + variability. In this work the variability in FC was studied by computing the + s.d. at various decades. An inverted U-shape was noted in quantifying the + variability in two of the three datasets where the middle-aged-to-elder subjects\u2014i.e. + 51\u201360 y.o.\u2014showed the highest variability. The findings point toward + brain aging being a complex process where alterations in cohorts do not simply + increase, decrease, or remain constant across the decades of life. The peak + in variability during middle age presents intriguing possibilities as far + as peoples\u2019 lifestyle, occupation, and perhaps their personality being + the most different during that juncture. Interestingly, the peak occurred + at different junctures within the middle-to-elder range for male and female + subjects.

\n

Another largely neglected facet of connectivity-based fMRI + analysis is the consideration of anticorrelations. While we note that the + relative change of significant anticorrelations during aging generally follow + what was seen for the positive correlations, there were discrepancies. Perhaps + more importantly, including the negative correlations in a tally of the connectivity + changes does lead to altered or new findings. With the 1000FCP subjects, if + one was to restrict attention to the canonical positive connections, then + a significant decrease in connectivity would be seen between the two earliest + decades of the considered ages. However, incorporating the anticorrelations + into the total indicates that the change is not significant. Additionally, + for the 1000FCP data, the loss of FC in the male subjects for the 81+ y.o. + group was not deemed significant when accounting for only the positive correlations + (or the anticorrelations)\u2013yet, the aggregate account shows statistical + significance. A similar result occurred for the oldest group of male SRPBS + subjects. Namely, counting the standard (positive) correlations showed no + significant alteration in the number of functional connections, whereas the + inclusion of anticorrelations indicated that the FC of 71+ y.o. males is modulated + in comparison to the reference young group. This reveals that when assessing + FC, an aggregate connectivity measure should be considered rather than solely + positive correlations.

\n

Human and animal studies are increasingly becoming + cognizant of the differences between the sexes and performing separate analyses. + Thus, the attainment of trajectories of healthy brain aging in males and females + is important. Repeating the per-decade FC analysis for male and female subjects + provided several findings. First, the changes in connectivity with aging were + similar between males and females in the three datasets. Second, the modulation + in the dynamics were more pronounced (i.e. statistically significant) for + males. Such a result would be important to take into account when assessing + deviations from a baseline in a male versus a female. Perhaps most importantly, + the results reveal that the modulation in FC with age is not as prominent + when male and female subjects are considered separately rather than together. + This accentuates the prudence of separate sex analyses in age-related studies + across a population. The inter-subject variability in FC among the sexes also + yielded interesting results. Although the peak variabilities occurred at different + junctures in the age spectrum between males and females, they occurred in + the middle-to-elder range. This result, however, did not hold in one dataset + where the youngest group of males exhibited the largest variance. Such discrepancy + highlights the general difficulty in attaining consensus across datasets in + light of the brain being a nonlinear stochastic system while aging is a complex + process. It also warrants the requirement of further scrutiny via additional + datasets, methodologies, and computation.

\n

The datasets considered + in this work share similarities as well as differences. The 1000FCP subjects + spanned 17 centers necessitating an investigation of site effects. The monotonicity + and change in FC analyses were repeated on subjects from the NKI-RS center + to investigate if the trends noted with all subjects coincide with those seen + for the largest constituent center. Interval 1 results on the presence and + degree of monotonicity coincided with those attained for 1000FCP. The interval + 2 findings also agreed on the absence of a monotonic relationship. Contrarily, + the interval 3 conclusions were different because the NKI-RS analysis did + not reflect a monotonic relationship. Aside from possible site effects, such + deviation may be due to the single-site dataset providing a fewer number of + subjects in the ML training set as well as in the test set. The change in + FC analysis conducted on the NKI-RS subjects provided results that were very + similar to the aggregate results. In the constituent and the 1000FCP, there + was an increase in variability among middle age to elder subjects, and a reduction + in FC variability for the two oldest decades. The trends noted for the positive + and negative correlations also coincided. Furthermore, the sex differences + in FC that were observed in the NKI-RS subjects agreed with those of the collective + dataset. In summary, the results are suggestive of site effects not being + prominent in the results presented for the 1000FCP recordings. Nevertheless, + the use of harmonization techniques [36, + 37] is a future + avenue for further assessing to what degree the presented results may be affected.

\n

The + SRPBS and camCAN studies maintained a constant protocol with respect to all + subjects having their eyes open and closed, respectively, during the recordings. + This was not the case with the 1000FCP dataset (S1 Table). It is not unusual for rsfMRI investigations + to consider a mixture of subjects of different arousal states [11, 38]. + Nevertheless, we consider the monotonicity and quantification of FC separately + for the 1000FCP recordings where subjects had their eyes open. The conclusions + of the monotonicity analysis when subjects had their eyes open (N = 543) generally + agreed with those attained with the aggregated subjects (S2 Fig). The hypotheses derived from interval + 1 and interval 2 agree between the two studies. However, the interval 3 results + indicate the absence of a monotonic relationship between the chronological + age and the percentage of subjects classified as aged when considering only + subjects with eyes open during recordings (S2C Fig). The findings pertaining to the + change in FC across decades were consistent when considering all 1000FCP subjects + and separately studying the subjects with eyes open (S2D Fig).

\n

The study of FC alterations + with aging was repeated with thresholds of |\u03C1 = 0.45| and |\u03C1 = 0.7| + for declaring connections (S3 + and S4 + Figs). Scrutiny with lower and higher thresholds aims to expurgate spurious + relationships that would arise with noise-dominated signals (i.e. low SNR + recordings). With the decreased value of |\u03C1 = 0.45|, the per-dataset + trends in the mean and variability of the number of FCs across the age spectrum + matched what was reported in Fig + 6. There were naturally a larger number of FCs per age group with the + lower threshold. Similarly, although a higher threshold lowered the number + of FCs in every age group, the trends in the mean and variability were consistent + to those observed in Fig + 6. The results are reassuring by reproducing the findings despite less + stringent as well as more conservative threshold values from what we initially + considered. Nevertheless, more detailed analysis of the FC dependency during + aging on the SNR and threshold values is an avenue of future research. There + is an additional caveat of the number of ROIs not being constant among the + considered datasets. However, it is not uncommon in fMRI studies that examine + multiple datasets to assess BOLD differences with different ROI numbers and + selection schemes [39\u201341]. For instance, the + authors of [42] + compared age prediction results with 419 ROIs from HCP subjects and 55 ROIs + from UK Biobank subjects. Despite the ML pipeline ensuring that the same number + of young and aged subjects are used to train an SVM, there is an unequal number + of subjects considered across the age groups in the datasets. This is a consequence + of the subject imbalance between ages that exists in neuroimaging studies + where age is a predictive variable and spans a broad range. Rather than being + a hinderance, this is standard among related works [8, 43]. + The number of data points across the subjects at different recording centers + was also not constant. This is consistent with related works such as [11, 44]. A recent work evaluated brain age + prediction accuracy for varying number of time points and observed improvements + with the number of time points [45]. + However, the increases were reported as being relatively small. Lastly, it + should be noted that more sophisticated and computationally demanding algorithmic + techniques such as DL may provide more accurate prediction results and reveal + FC properties in any one of the considered datasets. However, the techniques + may also overfit a different dataset to provide marginal predictive capability + and obscure FC characteristics by not generalizing. An SVM classifier was + considered in this work due to its simplicity, general robustness, and interpretability. + A comparative study with more advanced methods is a future avenue of investigation.

\n\n\nSupporting information\n\n\n\nBrain age + monotonicity without application of a sliding window.\n

The results + of the brain age monotonicity study for intervals 1, 2, and 3 without application + of a 5-year sliding window. To test for a trend, a linear regression was fit + to the data with the intercept \u03B20 and slope \u03B21. + A) For 1000FCP subjects, in interval 1 we have \u03B20 = 31.49, + \u03B21 = 0.971. The Spearman correlation of 0.549 computed among + the percentages of subjects classified as aged and the ages indicates a moderate + degree of monotonicity between the two quantities. In interval 2 we attain + \u03B20 = 70.94, \u03B21 = -0.108 indicating little + change in the classification of increasingly older brains as aged. The Spearman + correlation of -0.053 reflects a non-monotonic relationship among the ages + and the percentage of subjects classified as aged. For interval 3, the values + \u03B20 = 24.33, \u03B21 = 0.568, and the Spearman correlation + of 0.404 signifies a gradual increase and small degree of monotonicity. B) + When considering the SRPBS subjects, in interval 1 we have \u03B20 + = 6.647, \u03B21 = 2.749. The Spearman correlation of 0.818 computed + among the percentages of subjects classified as aged and the ages demonstrates + monotonicity between the two quantities. In interval 2 we attain \u03B20 + = 82.49, \u03B21 = 0.054 indicating little change in the classification + of increasingly older brains as aged. The Spearman correlation of 0.108 reflects + a non-monotonic relationship among the subjects\u2019 ages and the percentage + labeled as aged. For interval 3, the values \u03B20 = 3.48, \u03B21 + = 0.748, and the Spearman correlation of 0.544 indicates an increase and moderate + degree of monotonicity. C) For the camCAN dataset, in interval 1 we have \u03B20 + = 0, \u03B21 = 2.254. The Spearman correlation of 0.9027 computed + among the percentages of subjects classified as aged and the subject ages + signifies a high degree of monotonicity between the two quantities. In interval + 2 we attain \u03B20 = 76.42, \u03B21 = 0.3706 demonstrating + little change in the classification of increasingly older brains as aged. + The Spearman correlation of 0.386 reflects a non-monotonic relationship. For + interval 3, the values \u03B20 = 32.301, \u03B21 = -0.1968, + and the Spearman correlation of -0.118 indicate a gradual decrease and non-monotonicity + in the number of subjects classified as aged with increasing biological age. + D) With the NKI-RS subjects, in interval 1 we have \u03B20 = 39.71, + \u03B21 = 0.858. The Spearman correlation of 0.396 demonstrates + small to no monotonicity between the percentages of subjects classified as + aged and the ages. In interval 2 we attain \u03B20 = 64.78, \u03B21 + = 0.496 which indicates a gradual increase in the classification of increasingly + older brains as aged. The Spearman correlation of 0.371 reflects an absence + of monotonicity among the subjects\u2019 ages and the percentage labeled as + aged. For interval 3, the values \u03B20 = 45.75, \u03B21 + = -0.206, and the Spearman correlation of -0.112 indicates a non-monotonic + but decreasing relationship.

\n

(TIFF)

\n\n\n\n\n\nFindings + for subjects with eyes open during the rsfMRI recordings.\n

The + results of the brain age monotonicity and change in FC studies for subjects + in the 1000FCP dataset that had eyes open during rsfMRI recording. For the + monotonicity study, the histograms show the number of subjects classified + as young and aged at every considered age. A 5-year sliding window was applied + and a linear regression fit to the results with the intercept \u03B20 + and slope \u03B21. A) The analysis consists of a machine trained + on the youngest and oldest groups and presented with test data of intermediate-aged + subjects. In interval 1 we have \u03B20 = 22.71, \u03B21 + = 1.34. The Spearman correlation of 0.661 computed among the percentages of + subjects classified as aged and the ages indicates a moderate degree of monotonicity. + B) A machine was trained on the two youngest groups of subjects and then provided + with recordings from subjects 41+ y.o. (interval 2). The values \u03B20 + = 54.2, \u03B21 = 0.76, and a Spearman correlation of 0.325 indicate + an absence of monotonicity. C) A machine was trained on subjects from the + two oldest age groups prior to being presented with the task of classifying + younger subjects (21\u201360 y.o.) in interval 3. The linear fit values \u03B20 + = 34.32, \u03B21 = 0.25, and Spearman correlation of 0.047 signify + a non-monotonic relationship between chronological age and the percentage + of subjects classified as aged. D) The change in FC analysis consisted of + the mean FC computed by averaging across subjects the number of times that + the Pearson CCs from the connectivity matrix exceeded |\u03C1| + = 0.6. The s.d. of the number of functional connections is shown to study + the inter-subject variability across the age spectrum.

\n

(TIFF)

\n\n\n\n\n\nAnalysis + of FC changes with threshold |<italic>\u03C1</italic>| = 0.45.\n

A + study of the change in FC for subjects across decades from the A) 1000FCP, + B) SRPBS, and C) camCAN datasets. A threshold |\u03C1| = + 0.45 was used to determine if the Pearson CCs from the connectivity matrix + constituted a FC. The mean FC was computed by averaging the FC values across + the subjects in each age group. A bifurcation of the FC to positive and negative + (i.e. anticorrelations) directions is also shown. Two-sample t-tests were + used to assess the significance of the change in FC with the youngest age + group taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons of positive, negative, and total + FC values. The s.d. of the number of functional connections is illustrated + to study the inter-subject variability across the age spectrum.

\n

(TIFF)

\n\n\n\n\n\nAnalysis + of FC changes with threshold |<italic>\u03C1</italic>| = 0.7.\n

A + study of the change in FC for subjects across decades from the A) 1000FCP, + B) SRPBS, and C) camCAN datasets. A threshold |\u03C1| = + 0.7 was used to determine if the Pearson CCs from the connectivity matrix + constituted a FC. The mean FC was computed by averaging the FC values across + the subjects in each age group. A bifurcation of the FC to positive and negative + (i.e. anticorrelations) directions is also shown. Two-sample t-tests were + used to assess the significance of the change in FC with the youngest age + group taken as the reference. A p-value of 0.05 is the threshold for statistical + significance in the pairwise comparisons of positive, negative, and total + FC values. The s.d. of the number of functional connections is illustrated + to study the inter-subject variability across the age spectrum.

\n

(TIFF)

\n\n\n\n\n\n1000FCP + subjects with eyes open or closed.\n

A listing of the subjects from + the 1000FCP with eyes open (N = 543) or closed (N = 197) during the rsfMRI + recording.

\n

(TIFF)

\n\n\n\n\n\n\n

The + author acknowledges Itamar Kahn and Noam Bosak for providing the 1000FCP data, + material and methods, as well as feedback on the manuscript. I also thank + Daniela Kaufer and Jack Gallant for valuable discussions.

\n\n\nReferences\nKim + J., et al., \u201CAbnormal + intrinsic brain functional network dynamics in Parkinson\u2019s disease,\u201D + Brain, 2017. doi: 10.1093/brain/awx233 + 29053835\nHabes + M., et al., \u201CThe + Brain Chart of Aging: Machine-learning analytics reveals links between brain + aging, white matter disease, amyloid burden, and cognition in the iSTAGING + consortium of 10,216 harmonized MR scans,\u201D Alzheimer\u2019s + & Dementia, 2021. doi: 10.1002/alz.12178 + 32920988\nMattson + M. and Arumugam + T., \"Hallmarks of brain + aging: adaptive and pathological modification by metabolic states,\" + Cell Metabolism, 2018. doi: 10.1016/j.cmet.2018.05.011 29874566\nCole + J., et al., \u201CAbnormal + intrinsic brain functional network dynamics in Parkinson\u2019s disease,\u201D + Molecular Psychiatry, 2018.\nDosenbach + N., et al., \u201CPrediction + of individual brain maturity using fMRI,\u201D Science, + 2010. doi: 10.1126/science.1194144 20829489\nCole + J., \u201CMulti-modality + neuroimaging brain-age in UK Biobank: relationship to biomedical, lifestyle + and cognitive factors,\u201D Neurobiology of Aging, + 2020.\nKhosla + M., Jamison + K., Ngo + G., Kuceyeski + A., and Sabuncu + M., \u201CMachine learning + in resting-state fMRI analysis,\u201D Magnetic Resonance + Imaging, 2019. doi: 10.1016/j.mri.2019.05.031 + 31173849\nZhou + Z., et al., \u201CMultiscale + functional connectivity patterns of the aging brain learned from harmonized + rsfMRI data of the multi-cohort iSTAGING study,\u201D NeuroImage, + 2023. doi: 10.1016/j.neuroimage.2023.119911 + 36731813\nPervaiz + U., Vidaurre + D., Woolrich + M., and Smith + S., \u201COptimising network + modelling methods for fMRI,\u201D NeuroImage, + 2020. doi: 10.1016/j.neuroimage.2020.116604 + 32062083\nFair + D., et al., \u201CFunctional + brain networks develop from a \u201Clocal to distributed\u201D organization,\u201D + PLoS Computational Biology, 2009.\nSiman-Tov + T., et al., \u201CEarly + age-related functional connectivity decline in high-order cognitive networks,\u201D + Frontiers in Aging Neuroscience, 2017. doi: + 10.3389/fnagi.2016.00330 28119599\nAbdulrahman + H., et al., \u201CDopamine + and memory dedifferentiation in aging,\u201D NeuroImage, + 2017. doi: 10.1016/j.neuroimage.2015.03.031 + 25800211\nMennes + M., Biswal + B., Castellanos + F., and Milham + M., \u201CMaking data sharing + work: the FCP/INDI experience,\u201D NeuroImage, + 2013. doi: 10.1016/j.neuroimage.2012.10.064 + 23123682\nNooner + K., et al., \u201CThe + NKI-Rockland sample: a model for accelerating the pace of discovery science + in psychiatry,\u201D Frontiers in Neuroscience, + 2012.\nVan + Dijk K., et al., + \u201CIntrinsic functional connectivity as a tool for human + connectomics: theory, properties, and optimization,\u201D + Journal of Neurophysiology, 2010. doi: + 10.1152/jn.00783.2009 19889849\nKahn + I., et al., \u201CDistinct + cortical anatomy linked to subregions of the medial temporal lobe revealed + by intrinsic functional connectivity,\u201D Journal + of Neurophysiology, 2008. doi: 10.1152/jn.00077.2008 + 18385483\nTanaka + S., et al., \u201CA + multi-site, multi-disorder resting-state magnetic resonance image database,\u201D + Scientific Data, 2021. doi: 10.1038/s41597-021-01004-8 34462444\nShafto + M., et al., \u201CThe + Cambridge Centre for Ageing and Neuroscience (Cam-CAN) study protocol: a cross-sectional, + lifespan, multidisciplinary examination of healthy cognitive ageing,\u201D + BMC Neurology, 2014. doi: 10.1186/s12883-014-0204-1 25412575\nTaylor + J., et al., \u201CThe + Cambridge Centre for Ageing and Neuroscience (Cam-CAN) data repository: Structural + and functional MRI, MEG, and cognitive data from a cross-sectional adult lifespan + sample,\u201D NeuroImage, 2017. + doi: 10.1016/j.neuroimage.2015.09.018 + 26375206\nKahn + I. and Shohamy + D., \u201CIntrinsic connectivity + between the hippocampus, nucleus accumbens, and ventral tegmental area in + humans,\u201D Hippocampus, 2013. + doi: 10.1002/hipo.22077 23129267\nYahata + N., et al., \u201CA + small number of abnormal brain connections predicts adult autism spectrum + disorder,\u201D Nature Communication, 2016. + doi: 10.1038/ncomms11254 27075704\nPower + J., et al., \u201CMethods + to detect, characterize, and remove motion artifact in resting state fMRI,\u201D + NeuroImage, 2014. doi: 10.1016/j.neuroimage.2013.08.048 + 23994314\nAshburner + J. and Friston + K., \u201CUnified segmentation,\u201D + NeuroImage, 2005. doi: 10.1016/j.neuroimage.2005.02.018 + 15955494\nAshburner + J., \u201CA fast diffeomorphic + image registration algorithm,\u201D NeuroImage, + 2007. doi: 10.1016/j.neuroimage.2007.07.007 + 17761438\nTomasi + D. and Volkow + N., \u201CFunctional connectivity + density mapping,\u201D PNAS, 2010. + doi: 10.1073/pnas.1001414107 20457896\nVergun + S., et al., \u201CCharacterizing + functional connectivity differences in aging adults using machine learning + on resting state fMRI data,\u201D Frontiers in Computational + Neuroscience, 2013. doi: 10.3389/fncom.2013.00038 + 23630491\nHan + L., et al., \u201CFunctional + parcellation of the cerebral cortex across the human adult lifespan,\u201D + Cerebral Cortex, 2018. doi: 10.1093/cercor/bhy218 30307480\nHughes + C. et al., \u201CAging + relates to a disproportionately weaker functional architecture of brain networks + during rest and task states,\u201D NeuroImage, + 2020. doi: 10.1016/j.neuroimage.2020.116521 + 31926282\nSong + J., et al., \"Age-related + reorganizational changes in modularity and functional connectivity of human + brain networks,\" Brain Connectivity, 2014. + doi: 10.1089/brain.2014.0286 25183440\nMalagurski + B., et al., \u201CFunctional + dedifferentiation of associative resting state networks in older adults\u2013a + longitudinal study,\u201D NeuroImage, 2020. + doi: 10.1016/j.neuroimage.2020.116680 + 32105885\nDamoiseaux + J., \u201CEffects of aging + on functional and structural brain connectivity,\u201D Neuroimage, + 2017. doi: 10.1016/j.neuroimage.2017.01.077 + 28159687\nGeerligs + L., et al., \"A + brain-wide study of age-related changes in functional connectivity,\" + Cerebral Cortex, 2015. doi: 10.1093/cercor/bhu012 24532319\nLi + X., et al., \u201CAge-related + changes in brain structural covariance network,\u201D Frontiers + in Human Neuroscience, 2013.\nHrybouski + S., et al., \u201CInvestigating + the effects of healthy cognitive aging on brain functional connectivity using + 4.7 T resting-state functional magnetic resonance imaging,\u201D + Brain Structure and Function, 2021.\nOschmann + M., et al., \u201CA + longitudinal study of changes in resting-state functional magnetic resonance + imaging functional connectivity networks during healthy aging,\u201D + Brain Connectivity, 2020. doi: 10.1089/brain.2019.0724 32623915\nYu + M., et al., \u201CStatistical + harmonization corrects site effects in functional connectivity measurements + from multi-site fMRI data,\u201D Human Brain Mapping, + 2018. doi: 10.1002/hbm.24241 29962049\nPomponio + R., et al., \u201CHarmonization + of large MRI datasets for the analysis of brain imaging patterns throughout + the lifespan,\u201D NeuroImage, 2020. + doi: 10.1016/j.neuroimage.2019.116450 + 31821869\nRaut + R., et al., \u201COrganization + of propagated intrinsic brain activity in individual humans,\u201D + Cerebral Cortex, 2020. doi: 10.1093/cercor/bhz198 31504262\nGeerligs + L., et al., \u201CState + and trait components of functional connectivity: individual differences vary + with mental state,\u201D Journal of Neuroscience, + 2015. doi: 10.1523/JNEUROSCI.1324-15.2015 + 26468196\nDadi + K., et al., \u201CBenchmarking + functional connectome-based predictive models for resting-state fMRI,\u201D + NeuroImage, 2019. doi: 10.1016/j.neuroimage.2019.02.062 + 30836146\nAhrends + C., et al., \u201CData + and model considerations for estimating time-varying functional connectivity + in fMRI,\u201D NeuroImage, 2022. + doi: 10.1016/j.neuroimage.2022.119026 + 35217207\nHe T., + et al., \u201CDeep neural networks and kernel + regression achieve comparable accuracies for functional connectivity prediction + of behavior and demographics,\u201D NeuroImage, + 2020. doi: 10.1016/j.neuroimage.2019.116276 + 31610298\nKim + E., et al., \u201CConnectome-based + predictive models using resting-state fMRI for studying brain aging,\u201D + Experimental Brain Research, 2022. doi: + 10.1007/s00221-022-06430-7 35922524\nHaak + K., Marquand + A., and Beckmann + C., \u201CConnectopic mapping + with resting-state fMRI,\u201D NeuroImage, + 2018. doi: 10.1016/j.neuroimage.2017.06.075 + 28666880\nGuan + S., Jiang + R., Meng + C., and Biswal + B., \u201CBrain age prediction + across the human lifespan using multimodal MRI data,\u201D + GeroScience, 2024. doi: 10.1007/s11357-023-00924-0 + 37733220\n\n\n
" + headers: + Accept-Ranges: + - bytes + Cache-Control: + - private, max-age=0 + Content-Disposition: + - attachment; filename=journal.pone.0300720.xml + Content-Length: + - '148585' + Content-Type: + - application/xml + Date: + - Wed, 23 Sep 2026 16:31:41 GMT + ETag: + - '"4631075318eab9203f6900b736a41fb7"' + Expires: + - Wed, 23 Sep 2026 16:31:41 GMT + Last-Modified: + - Thu, 30 May 2024 17:34:49 GMT + Server: + - UploadServer + X-GUploader-UploadID: + - AJjja9b39Uf0wddp1OKdgWUkVi-XiE86l3VfruRYVxXdoGAqiCzvV0S81RFld4p2vqT7G-xD + x-goog-generation: + - '1717090489408886' + x-goog-hash: + - crc32c=6gBZlA== + - md5=RjEHUxjquSA/aQC3NqQftw== + x-goog-metageneration: + - '1' + x-goog-storage-class: + - STANDARD + x-goog-stored-content-encoding: + - identity + x-goog-stored-content-length: + - '148585' + status: + code: 200 + message: OK +- request: + body: null + headers: + Accept: + - '*/*' + Accept-Encoding: + - gzip, deflate + Connection: + - keep-alive + User-Agent: + - Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like + Gecko) Chrome/124.0.6367.155 Safari/537.36 + method: GET + uri: https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0303969&type=manuscript + response: + body: + string: '' + headers: + Alt-Svc: + - h3=":443"; ma=2592000,h3-29=":443"; ma=2592000 + Cache-Control: + - max-age=3600 + Content-Length: + - '0' + Date: + - Wed, 23 Sep 2026 16:31:42 GMT + Location: + - https://storage.googleapis.com/plos-corpus-prod/10.1371/journal.pone.0303969/1/pone.0303969.xml?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=wombat-sa%40plos-prod.iam.gserviceaccount.com%2F20260923%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20260923T163011Z&X-Goog-Expires=86400&X-Goog-SignedHeaders=host&X-Goog-Signature=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 + Server: + - Apache + Set-Cookie: + - GCLB=CImT2u_AqN-MARAD; path=/; HttpOnly + Strict-Transport-Security: + - max-age=31536000 + - max-age=31536000 ; includeSubDomains + Via: + - 1.1 google + X-Content-Type-Options: + - nosniff + X-Frame-Options: + - DENY + X-XSS-Protection: + - 1; mode=block + status: + code: 302 + message: Found +- request: + body: null + headers: + Accept: + - '*/*' + Accept-Encoding: + - gzip, deflate + Connection: + - keep-alive + User-Agent: + - Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like + Gecko) Chrome/124.0.6367.155 Safari/537.36 + method: GET + uri: https://storage.googleapis.com/plos-corpus-prod/10.1371/journal.pone.0303969/1/pone.0303969.xml?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=wombat-sa%40plos-prod.iam.gserviceaccount.com%2F20260923%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20260923T163011Z&X-Goog-Expires=86400&X-Goog-SignedHeaders=host&X-Goog-Signature=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 + response: + body: + string: "\n\n
\n\n\nPLoS ONE\nplos\nplosone\n\nPLOS + ONE\n\n1932-6203\n\nPublic + Library of Science\nSan Francisco, CA USA\n\n\n\n10.1371/journal.pone.0303969\nPONE-D-22-29235\n\n\nResearch Article\n\n\nPhysical sciencesChemistryChemical + elementsGadolinium\nMedicine and health sciencesCardiologyHeart + rate\nBiology + and life sciencesDevelopmental biologyFibrosis\nPeople and placesPopulation + groupingsAge groups\nPhysical sciencesPhysicsRelaxation + (physics)Relaxation time\nBiology and life sciencesAnatomyCardiovascular + anatomyHeartMyocardium\nMedicine and health sciencesAnatomyCardiovascular + anatomyHeartMyocardium\nMedicine and health sciencesMedical + conditionsNeurodegenerative diseasesMovement + disordersFriedreich's ataxia\nMedicine and health sciencesNeurologyNeurodegenerative + diseasesMovement disordersFriedreich's + ataxia\nPhysical sciencesPhysicsCondensed + matter physicsMagnetismMagnetic + resonance\n\nInsights + into the effects of Friedreich ataxia on the left ventricle using T1 mapping + and late gadolinium enhancement\nLeft + ventricular effects of Friedreich ataxia\n\n\n\nhttps://orcid.org/0000-0002-3319-5356\n\nPeverill\nRoger E.\n\nConceptualization\nData + curation\nFormal + analysis\nFunding + acquisition\nInvestigation\nMethodology\nProject + administration\nResources\nSoftware\nSupervision\nValidation\nVisualization\nWriting + \u2013 original draft\nWriting + \u2013 review & editing\n1\n*\n\n\n\nLin\nKimberly + Y.\n\nConceptualization\nData + curation\nFormal + analysis\nFunding + acquisition\nInvestigation\nMethodology\nProject + administration\nResources\nSoftware\nSupervision\nValidation\nWriting + \u2013 review & editing\n2\n\n\n\nFogel\nMark + A.\n\nConceptualization\nData + curation\nFormal + analysis\nFunding + acquisition\nInvestigation\nMethodology\nProject + administration\nResources\nSoftware\nSupervision\nValidation\nWriting + \u2013 review & editing\n2\n\n\n\nCheung\nMichael + M. H.\n\nData + curation\nFormal + analysis\nFunding + acquisition\nInvestigation\nMethodology\nProject + administration\nResources\nSoftware\nSupervision\nValidation\nWriting + \u2013 review & editing\n3\n4\n5\n\n\n\nMoir\nW. + Stuart\n\nData + curation\nFormal + analysis\nInvestigation\nMethodology\nResources\nSoftware\nSupervision\nValidation\nWriting + \u2013 review & editing\n1\n\n\n\nCorben\nLouise + A.\n\nData + curation\nInvestigation\nProject + administration\nResources\nWriting + \u2013 review & editing\n6\n\n\n\nCahoon\nGlenn\n\nMethodology\nWriting + \u2013 review & editing\n3\n\n\n\nDelatycki\nMartin + B.\n\nConceptualization\nFunding + acquisition\nProject + administration\nResources\nSupervision\nWriting + \u2013 review & editing\n6\n7\n\n\n Monash Cardiovascular Research Centre, + MonashHeart and Department of Medicine (School of Clinical Sciences at Monash + Health), Monash University and Monash Health, Clayton, Victoria, Australia\n Division of Cardiology, Children\u2019s + Hospital of Philadelphia, Philadelphia, Pennsylvania, United States of America\n Department of Cardiology, Royal + Children\u2019s Hospital, Parkville, Victoria, Australia\n Department of Paediatrics, University + of Melbourne, Parkville, Victoria, Australia\n + Heart Research Group, Murdoch Children\u2019s Research Institute, + Parkville, Victoria, Australia\n + Bruce Lefroy Centre for Genetic Health Research, Murdoch Children\u2019s + Research Institute, Parkville, Victoria, Australia\n Victorian Clinical Genetics Services, + Parkville, Victoria, Australia\n\n\n\nOjha\nVineeta\n\nEditor\n\n\n\nAll + India Institute of Medical Sciences, INDIA\n\n\n

The authors have declared that no + competing interests exist.

\n\n* E-mail: roger.peverill@monash.edu\n\n\n30\n5\n2024\n\n\n2024\n\n19\n5\ne0303969\n\n\n23\n10\n2022\n\n\n4\n5\n2024\n\n\n\n2024\nPeverill + et al\n\nThis is an open access article distributed + under the terms of the Creative Commons Attribution License, which + permits unrestricted use, distribution, and reproduction in any medium, provided + the original author and source are credited.\n\n\n\n\n\nBackground\n

The left ventricular (LV) changes + which occur in Friedreich ataxia (FRDA) are incompletely understood.

\n\n\nMethods\n

Cardiac magnetic resonance (CMR) + imaging was performed using a 1.5T scanner in subjects with FRDA who are homozygous + for an expansion of an intron 1 GAA repeat in the FXN gene. + Standard measurements were performed of LV mass (LVM), LV end-diastolic volume + (LVEDV) and LV ejection fraction (LVEF). Native T1 relaxation time and the + extracellular volume fraction (ECV) were utilised as markers of left ventricular + (LV) diffuse myocardial fibrosis and late gadolinium enhancement (LGE) was + utilised as a marker of LV replacement fibrosis. FRDA genetic severity was + assessed using the shorter FXN GAA repeat length (GAA1).

\n\n\nResults\n

There were 93 subjects with FRDA + (63 adults, 30 children, 54% males), 9 of whom had a reduced LVEF (<55%). + A LVEDV below the normal range was present in 39%, a LVM above the normal + range in 22%, and an increased LVM/LVEDV ratio in 89% subjects. In adults + with a normal LVEF, there was an independent positive correlation of LVM with + GAA1, and a negative correlation with age, but no similar relationships were + seen in children. GAA1 was positively correlated with native T1 time in both + adults and children, and with ECV in adults, all these associations independent + of LVM and LVEDV. LGE was present in 21% of subjects, including both adults + and children, and subjects with and without a reduced LVEF. None of GAA1, + LVM or LVEDV were predictors of LGE.

\n\n\nConclusion\n

An + association between diffuse interstitial LV myocardial fibrosis and genetic + severity in FRDA was present independently of FRDA-related LV structural changes. + Localised replacement fibrosis was found in a minority of subjects with FRDA + and was not associated with LV structural change or FRDA genetic severity + in subjects with a normal LVEF.

\n\n\n\n\n\n\nhttp://dx.doi.org/10.13039/100002108\nFriedreich's + Ataxia Research Alliance\n\n\nBronya + J Keats International Research Collaboration Award\n\n\nDelatycki\nMartin + B.\n\n\n\nThis + study was funded by the Friedreich Ataxia Research Alliance (FARA) Bronya + J Keats International Research Collaboration Award (curefa.org). The funders + had an advisory role in study design and data collection, but no role in data + analysis, decision to publish or preparation of the manuscript.\n\n\n\n\n\n\n\n\nData Availability\nAll + relevant data are within the manuscript and its Supporting + information files.\n\n\n\n\n\n\nIntroduction\n

Friedreich + ataxia (FRDA) is an autosomal recessive neurodegenerative disease caused by + pathogenic variants in the FXN gene which encodes for the + mitochondrial protein frataxin [1]. + Cardiac disease is a frequent accompaniment of FRDA, can lead to arrhythmias + and cardiac failure, and is the most common cause of death [2\u20134]. + A proportion of subjects with FRDA develop a reduced LV ejection fraction + (LVEF) [5\u20137], and post mortem histology + of the heart in FRDA has shown extensive interstitial fibrosis, focal degeneration + of muscle fibres and active muscle necrosis [8]. + An increase in LV wall thickness and a reduction in LV chamber size are more + common than, and likely precede and predispose to, a subsequent reduction + of LVEF in FRDA [5\u20137, 9\u201315]. + However, there is little information about the histological changes in the + LV myocardium during the early stages of the disease process.

\n

A better + understanding of the changes which occur in the myocardium in FRDA prior to + a reduction in LVEF is important for future attempts to modify or prevent + cardiac disease progression. Cardiac magnetic resonance (CMR) not only provides + a gold standard assessment for LV volumes and LV mass (LVM), but with the + use of intravenous gadolinium and late imaging to detect late gadolinium enhancement + (LGE), it also provides a non-invasive technique for the visualisation of + localised replacement myocardial fibrosis [16]. + Furthermore, sensitive quantitative information about more generalised changes + in the LV myocardium can be obtained using T1 mapping techniques [17]. Both native T1 relaxation time and + calculated myocardial extracellular volume fraction (ECV) have been shown + to increase in the presence of diffuse myocardial fibrosis [17\u201321]. + However, there has only been limited investigation in FRDA into the presence + of LGE [6, 22\u201324] or diffuse myocardial fibrosis [24].

\n

Approximately + 96% of FRDA is due to homozygosity for a GAA expansion in intron 1 of FXN, + with the other 4% being compound heterozygous for a GAA expansion and a different + FXN pathogenic variant. In the homozygous group, the number + of GAA repeats in the shorter allele of the FXN gene (GAA1) + is inversely related to cellular levels of frataxin [25, 26], + and there has therefore been interest in the ability of GAA1 to explain cardiac + disease severity in FRDA. GAA1 has been reported to be a predictor of cardiac + death [3, 4], of progression to + a reduced LVEF [7], + and has also been associated with increased LV wall thickness [4, 9, + 10, 12, 15, + 27], relative wall + thickness (RWT) [22], + LV mass (LVM) [10, + 15, 28] and LVM index (LVMI) [9, 29] + and a smaller LV end-diastolic diameter (LVEDD) [4, + 15] and LV end-diastolic + volume (LVEDV) [14, + 15]. The aims of + the present study using CMR with gadolinium injection in adults and children + with FRDA homozygous for GAA repeat expansions, were to quantify LV volumes, + LVM, native T1 time and ECV, to determine the presence and extent of LGE, + and to investigate the relationship of these LV variables with GAA1, age at + onset of symptoms (AOS), symptom duration (SDur), age and the age group (i.e. + children versus adults).

\n\n\nMaterials + and methods\n\nSubjects\n

Children + and adults homozygous for a GAA expansion in intron 1 of FXN + and aged between 10 and 50 years were recruited through clinical research + programs (including the Collaborative Clinical Research Network in FA (CCRN-FA) + and from the Friedreich Ataxia clinics at Monash Health, Melbourne, Victoria, + Australia and the Children\u2019s Hospital of Philadelphia, PA, USA. Prior + to undergoing CMR all individuals had their renal function evaluated and estimated + glomerular filtration rate (eGFR) calculated to confirm the absence of renal + dysfunction and thus the safety of gadolinium administration. Individuals + with known ischemic heart disease, hypertension, more than mild valvular disease, + a persistent atrial arrhythmia or contraindications to CMR or gadolinium injection + were excluded from the study. Approval for this study was provided by the + Human Research Ethics Committee of the Royal Children\u2019s Hospital, Melbourne, + Victoria, Australia and the Institutional Review Board of the Children\u2019s + Hospital of Philadelphia, PA, USA. All participants in the study, or their + parents/guardians if the participant was aged under 18 years, provided written + informed consent as per the Declaration of Helsinki.

\n

Height and weight + were measured, and body surface area (BSA) and body mass index were calculated, + using standard formulae. AOS was defined as the time when an individual or + their parents first noted neurological or non-neurological symptoms of FRDA, + and SDur was defined as the time between AOS and the time of CMR testing. + GAA repeat size in the smaller (GAA1) and larger (GAA2) alleles of the FXN + gene were determined using a single in-house method at the Melbourne site + and from different commercial laboratories at the Philadelphia site. Clinical + severity of neurological disease was evaluated to provide a neurological profile + of the cohort by performance of the neurological component of the Friedreich + Ataxia Rating Scale (nFARS) within one month of the CMR study (available in + 81 subjects, score out of 125). A blood sample was drawn on the same day as + the CMR study for measurement of haematocrit (Hct).

\n\n\nCardiac + magnetic resonance protocol\n

CMR studies were performed using similar + protocols on 1.5T magnetic resonance imaging units (Magnetom Avanto in Philadelphia + and Magnetom Aera in Melbourne; Siemens Medical Solutions, Erlangen, Germany). + Breath-held images were acquired in expiration, and a 4-lead vector-electrocardiogram + was recorded for gating purposes. Following standard localising images, true + fast imaging with steady-state precession (TrueFISP) cine images were acquired + in the 4-chamber, 2-chamber, and 3-chamber views and used to plan short axis + imaging. End-expiration basal, mid-cavity, and apical short axis native T1 + images were collected using the optimised parameter set of a Modified Look-Locker + Inversion recovery (MOLLI) research sequence provided by Siemens with an acquisition + rest/schema of 5(3)3 designed for long T1 values in native scans.

\n

An + IV injection of gadobutrol (Gadovist 1.0, Bayer AG, Berlin, Germany) at a + dose of 0.2 mmol/kg body weight was administered to all subjects and immediately + following contrast injection, TrueFISP cine functional imaging was obtained + in contiguous 7.0 mm slices in the short axis. At 10 minutes post contrast + administration a TI scout sequence was used to determine optimal inversion + time (average TI = 300 ms) for the late gadolinium enhanced (LGE) images, + 8 mm single slice mid ventricle, TR/TE: 23.67/1.1 ms, flip angle 30\xB0, 340 + mm FOV, and generated using 20 ms increments from 80 to 500 ms. LGE images + were collected using a free breathing phase sensitive inversion recovery (PSIR) + technique, as well as single slice, breath held, PSIR with whole ventricular + coverage in the short axis. Factors for the two sequences were adjusted to + allow direct comparison, typically TR 700 ms, FOV 340 x 280 mm, flip angle + 45\xB0, matrix = 256 x 192, TI = 300 ms, voxel size 1.3 x 1.3 x 8.0 mm.

\n

Post + contrast T1 mapping sequences were acquired at approximately 15 minutes post + contrast injection using an optimised parameter set for MOLLI using the same + slice thickness and positions as the native T1 images. TR/TE = 360/112 ms, + flip angle = 35\xB0, TI = 260 ms, FOV = 360 x 306 mm, matrix = 256 x 168, + interpolated voxel size = 1.4 x 1.4 x 8.0 mm, GRAPPA = 2 with 36 reference + lines. An acquisition/rest schema of 4(1)3(1)2 was employed to account for + short TI values post contrast.

\n

Standard measurements of CMR volumes, + LVM and LVEF, and analysis of T1 maps, were performed at each site using cvi42 + v5.3.2 (Circle Cardiovascular Imaging Inc, Calgary, AB, Canada). Papillary + muscles were included in the myocardial mass and excluded from LV volumes. + The presence of abnormalities of LV end-diastolic volume (LVEDV) and LVM were + determined using sex and BSA based criteria specific for children or adults + as appropriate, with papillary muscles excluded from LVEDV and included in + LVM, and 2 standard deviations (SD) from the mean representing the upper and + lower limits of the normal range [30]. + Left ventricular hypertrophy (LVH) was defined as a LVM > 2 SD above the + mean. A LVEF of <55% was defined as reduced. The LVM volume ratio (LVMVR) + was calculated as the ratio of LVM/LVEDV and used as a marker of concentric + LV geometry, with an upper limit of the normal range of 0.9 in males and 0.8 + in females [30]. + For myocardial T1 analysis, manual epicardial and endocardial contours were + drawn on the MOLLI mid ventricular short axis slice and segmented according + to the American Heart Association (AHA) 16 segment model.

\n

Reported + T1 based measurements were an average of results from all the basal and mid + LV segments as has been previously described [31]. + Extracellular volume fraction (ECV) was calculated using the mean segmental + pixel value from the MOLLI ECV maps using the formula: ECV = (\u0394[1/T1myo]/\u0394[1/T1blood])*[1-Hct]). + A \"synthetic ECV\" was also calculated using a calculated haematocrit utilising + the equation: synthetic haematocrit = 831.6 * (1 / T1blood)\u20140.151 + [32]. Native T1 + times and ECV in individual FRDA subjects could not be classed into normal + or abnormal in this study because of the lack of an adequate comparison group. + This is because native T1 times and ECV are not only recognised to be machine + and protocol dependent, but also because they are sex- and probably also age-dependent + in adults [33\u201336]. Normal T1 values + therefore require large numbers of subjects of different ages and of both + sexes, but these were not available for either adults or children for this + study. To determine inter-rater reliability of T1 times in CMR studies, 19 + randomly chosen scans had T1 measurements repeated at the alternative site, + with the study images sent to the alternative site, and the second measurer + blinded to the previous measurements.

\n\n\nStatistical + analysis\n

Results are expressed as mean \xB1 standard deviation + or median [range] if not normally distributed. Statistical analysis has been + performed using Systat V13 (Systat Software, Chicago, IL, USA). Some of the + analyses have been confined to subjects with a normal LVEF because the number + of subjects with a reduced LVEF was small, and also because the main focus + of this study was on the early myocardial changes which occur in FRDA. Children + and adults were compared but also analysed separately because of previous + evidence that there may be differences in the relationship of GAA1 with LV + structural features based on age group [15]. + Univariate linear regression analysis was performed to assess the relationships + of CMR variables with selected demographic and FRDA-specific variables of + interest based on previous findings about the cardiac effects of FRDA. In + particular, BSA, sex, age, age group (child v adult), AOS, SDur and GAA1 were + considered in analyses of LV volumes and LVM, and sex [33\u201338], + age [33\u201336], age group, heart + rate [34, 38], AOS, SDur and GAA1 + were considered in analyses of T1 mapping variables. The relationship of GAA2 + with the above independent variables was also investigated but it was not + a predictor of any of the variables and has not been reported in any of the + analyses. Categoric variables were included in multivariate models as dummy + variables for sex (male = 1, female = 0), age group (adults = 1, children + = 0) and study site (Philadelphia = 1, Melbourne = 0). SDur was not normally + distributed and was log transformed prior to inclusion in linear regression + models. Log SDur was positively correlated with age (r = 0.48, p<0.001) + and GAA1 was negatively correlated with AOS (r = -0.58, p<0.001), and therefore + these relationships were considered when constructing multivariate models. + Independent variables were removed during the multivariate modelling process + if the p value was >0.10, but the final models only show independent variables + with a p<0.05. Scatter plots and residuals were reviewed to determine if + there were outliers, and if identified, the possibility of exclusion of the + outlier from the analysis was considered.

\n\n\n\nResults\n

There were 96 subjects with + FRDA who underwent CMR imaging, but of these, satisfactory LV volume and mass + measurements were not possible in 3 subjects, 2 because of an irregular heart + rate during image acquisition, and one because they were found to have been + dehydrated at the time of the study. The demographics of the remaining 93 + subjects, comprising 63 adults and 30 children, are shown in Table 1. There were three subjects with diabetes, + all of whom were adults. The numbers for Melbourne and Philadelphia are shown + separately and demonstrate some variations in subject demography between the + two sites.

\n\n10.1371/journal.pone.0303969.t001\n Demography of study subjects at each study site.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
\nTotal (n = 93)Philadelphia (n + = 56)Melbourne (n = 37)
Male50 (53.8%)28 + (50.0%)22 (59.5%)
Adults63 (67.7%)33 (58.9%)30 (80.6%)
Age (years)24.1\xB19.820.8\xB17.429.2\xB110.9
AOS (years)11.6\xB15.99.8\xB13.914.3\xB17.4
SDur (years)8.0 [0\u201333.8]5.0 [0.8\u201326.0]13.4 [0\u201333.8]
nFARS + score (n = 81)59\xB12063\xB11954\xB120
GAA1689\xB1223732\xB1228623\xB1213
GAA2921\xB1181937\xB1188897\xB1168
\n\n\n

AOS\u2014age at onset of symptoms; SDur\u2014symptom duration; + nFARS\u2014neurological Friedreich ataxia rating scale score; GAA1 \u2013number + of GAA repeats in the smaller allele of the FXN gene; GAA2\u2014number + of GAA repeats in the larger allele of the FXN gene

\n\n\n\nLeft ventricular mass and volumes\n

Of the + 93 subjects there were 20 (22%) with LVH, this comprising 10/30 (33%) of the + children and 10/63 (16%) of the adults. There were 36 subjects (39%) with + a small LVEDV, and this comprised 30% of the children and 42% of the adults. + There were only 3 subjects who met the criteria for a LVEDV above the normal + range, all of whom were adults. There were 9 subjects with a LVEF <55% + (6 adults, 3 children), of whom 2 also had LV dilatation and 3 also had an + increased LVM. The results of indexed LV volumes, LVMI, LVMVR and LVEF in + subjects with a LVEF \u226555% (n = 84) are shown in Table 2, with male and female subjects shown + separately. In a comparison of adult subjects with a normal LVEF, LVMI, LVEDVI + and LVESVI were all higher in males than females, whereas LVEF was higher + in females. LVMVR was greater than the upper limit of the sex-specific normal + range in 75/84 (89%) subjects, was higher in children than adults (1.47\xB10.33 + v 1.24\xB10.38 g/mL, p<0.01), but was similar in adult males and females + (p = 0.79).

\n\n10.1371/journal.pone.0303969.t002\n Left ventricular indexed volumes and mass in males + and females with a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
\nMalesFemales
N45 (32 adults)39 + (31 adults)
Age (years)25.6\xB110.722.0\xB18.0
Body surface + area (m2)1.79\xB10.33**1.58\xB10.24
Body mass index (kg/m2)22.8\xB14.722.0\xB16.3
LVEDVI + (mL/m2)64\xB113*57\xB112
LVESVI (mL/m2)18\xB18*14\xB17
Stroke volume index (mL/m2)46\xB1843\xB18
LVMI (g/m2)83\xB121**70\xB115
LVMVR (g/mL)1.36\xB10.401.28\xB10.35
LVEF (%)72\xB18*76\xB18
Heart rate (/min)81\xB11486\xB114
Cardiac index + (L/min/m2)3.65\xB10.653.69\xB10.95
\n\n\n

LVEDVI\u2014left ventricular end-diastolic volume index; + LVESVI\u2014left ventricular end-systolic volume index; LVMI\u2014left ventricular + mass index; LVMVR\u2014left ventricular mass volume ratio; LVEF\u2014left + ventricular ejection fraction

\n

*p<0.02 + and

\n

**p<0.01 for comparison of adult + males and females

\n\n\n\n\nPredictors of LVEDV in subjects with a normal LVEF\n

In + subjects with a LVEF \u226555% (n = 84), LVEDV was positively correlated with + BSA (r = 0.76, p<0.001) and was higher in males (p<0.001), but male + sex was only a borderline significant predictor of LVEDV after adjusting for + BSA (p = 0.053). After adjustment for BSA and sex, LVEDV was not different + between children and adults (p = 0.70). In adults (n = 57) after adjusting + for BSA and sex, AOS was a borderline positive correlate of LVEDV (p = 0.056), + but there were no contributions to the model of LVDEV from GAA1 (p = 0.39), + log SDur (p = 0.66) or age (p = 0.80). In a multivariate model of LVEDV, BSA + and AOS, but not sex, were independent predictors of LVEDV, with an earlier + AOS associated with a smaller LVEDV (Table + 3). In children (n = 27) BSA was a positive correlate of LVEDV (r = + 0.73, p<0.001), and after adjusting for BSA there was no independent contribution + from sex to the model (p = 0.80). After including BSA in multivariate models + of LVEDV in children there were no contributions from AOS (p = 0.17), GAA1 + (p = 0.12) or age (p = 0.15), but there was a borderline significant contribution + from log SDur (p = 0.05), with longer disease duration associated with a smaller + LVEDV.

\n\n10.1371/journal.pone.0303969.t003\n Multivariate model of LVEDV in adult subjects with + a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variabler value in + univariate analysis\u03B2 in + multivariate modelp value in multivariate modelCumulative adjusted r2
BSA0.670.58<0.0010.44
Male sex0.47\nNS0.47
AOS0.460.250.0190.49
\n\n\n

BSA\u2014body surface area; AOS\u2014age at the onset + of symptoms

\n\n\n\n\nPredictors + of LVM in subjects with a normal LVEF\n

In subjects with a LVEF + \u226555% (n = 84), LVM was positively correlated with BSA (p<0.001), larger + in males (p<0.001), and both BSA and sex were independent predictors of + LVM (p<0.001 for both). After adjustment for BSA and sex, LVM was not different + between children and adults (p = 0.11). In adults (n = 57) after including + BSA and sex in the model, there were additional contributions from GAA1 (\u03B2 + = 0.36), AOS (\u03B2 = -0.35) and age (\u03B2 = -0.39), but not from log SDur + (p = 0.57). The model which included GAA1 (Table + 4) explained a similar amount of the variance of LVM to the model which + contained AOS (Table 5). + Neither GAA1 nor AOS remained significant predictors in the model of LVM when + both independent variables were included together (p = 0.056 & p = 0.062, + respectively). Older age was an independent predictor of a lower LVM in a + model which included BSA, sex and GAA1 (Table + 4), whereas neither AOS nor age remained significant when age was added + to the combination of BSA, sex and AOS (p = 0.067 & p = 0.054, respectively). + GAA1 accounted for 10% and age accounted for an additional 5% of the variance + in LVM in adults. All the predictors of LVM in Table + 4 remained significant after excluding diabetic subjects from the analysis. + In children (n = 27) LVM was positively correlated with BSA (r = 0.54), and + male sex was a borderline significant predictor of a larger LVM (p = 0.066) + after adjusting for BSA. There were no significant contributions to the model + of LVM in children which included BSA and sex from any of GAA1 (p = 0.86), + AOS (p = 0.97), log SDur (p = 0.79) or age (p = 0.20).

\n\n10.1371/journal.pone.0303969.t004\n Multivariate model of LVM including GAA1 in adult + subjects with a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variabler value in + univariate analysis\u03B2 in multivariate modelp value in multivariate modelCumulative + adjusted r2
BSA0.620.55<0.0010.37
Male sex0.480.42<0.0010.42
GAA10.110.280.0080.52
Age-0.10-0.270.0070.57
\n\n\n

BSA\u2014body surface area, GAA1\u2014number of GAA repeats + in the smaller allele of the FXN gene

\n\n\n\n10.1371/journal.pone.0303969.t005\n Multivariate model of LVM including AOS in adult + subjects with a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variabler value in + univariate analysis\u03B2 in multivariate modelp value in multivariate modelCumulative + adjusted r2
BSA0.620.59<0.0010.37
Male sex0.480.36<0.0010.42
AOS-0.01-0.350.0010.51
\n\n\n

BSA\u2014body surface area; AOS\u2014age at the onset + of symptoms

\n\n\n\n\nT1 + relaxation times\n

T1 relaxation times pre and post gadolinium contrast, + the partition coefficient, and the ECV calculated using both the measured + Hct and the calculated Hct, in the total group, and subjects with normal and + reduced LVEF, are shown in Table + 6. Post contrast T1 imaging was not performed, and thus ECV was also + not available, in one of the subjects with a reduced LVEF. In the 87 subjects + in whom both methods for ECV calculation could be performed, the synthetic + ECV using the calculated Hct gave a slightly higher result (0.286\xB10.050 + v 0.277\xB10.051, p<0.001), but there was a close correlation between the + two methods (r = 0.97, p<0.001). Because it was available in all subjects + the synthetic ECV results have been used in all the modelling of ECV. Univariate + p values for comparison of normal LVEF with the reduced LVEF group are not + provided in Table 6 + because there were a number of potential confounding factors including sex + and study site (see below).

\n\n10.1371/journal.pone.0303969.t006\n T1 relaxation times.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
\nTotal groupLVEF\u226555%LVEF<55%
Native + T1 relaxation time (ms) (n = 93)1012\xB1421010\xB144 (n = 84)1019\xB146 + (n = 9)
Native blood T1 time (ms) (n + = 93)1569\xB1911564\xB191 + (n = 84)1606\xB193 (n = 9)
Post contrast T1 time (ms) (n = 92)476\xB164445\xB164 (n = 84)415\xB160 (n + = 8)
Post contrast blood T1 time (ms) + (n = 92)314\xB177316\xB180 + (n = 84)300\xB156 (n = 8)
Partition coefficient (n = 92)0.47\xB10.090.47\xB10.07 (n = 84)0.53\xB10.13 + (n = 8)
Synthetic ECV (n = 92)0.288\xB10.0510.284\xB10.045 + (n = 84)0.328\xB10.085 (n = 8)
ECV (n = 87)0.277\xB10.0510.274\xB10.046 (n = 80)0.314\xB10.084 + (n = 7)
\n\n\n

ECV\u2014extracellular volume fraction; Hct\u2014haematocrit

\n\n\n

In + univariate analyses of the whole group, native T1 time was not different between + the study sites (p = 0.70), but it was higher in women than men (p<0.001) + and positively correlated with heart rate (r = 0.30, p = 0.003). After adjusting + for sex and heart rate, there was no difference in native T1 time between + subjects with a reduced versus a normal LVEF (p = 0.17). Post contrast T1 + time and ECV were both higher at the Philadelphia than the Melbourne site + (p<0.002 for both). After adjusting for study site, post contrast T1 time + was higher in men (p = 0.047), but not correlated with heart rate (p = 0.30) + and not different between those with a normal versus reduced LVEF (p = 0.47). + After adjusting for study site, ECV was not correlated with heart rate (p + = 0.15), there was a trend for it to be higher in females (p = 0.094), and + it was higher in subjects with a reduced LVEF (p = 0.001).

\n\n\nPredictors of native T1 relaxation time in subjects + with a normal LVEF\n

In subjects with a normal LVEF (n = 84), after + adjusting for sex, heart rate was a positive correlate of native T1 time (\u03B2 + = 0.21, p = 0.029) and children had a higher native T1 time than adults (p + = 0.02), but when included together in a model of native T1 time, neither + heart rate (p = 0.09) nor age group (p = 0.061) remained significant predictors. + In adults with a normal LVEF (n = 57), after adjusting for sex, GAA1 was an + independent positive correlate of native T1 time (\u03B2 = 0.21, p = 0.013), + whereas there were no contributions from any of heart rate (p = 0.30), AOS + (p = 0.57), log SDur (p = 0.35), age (p = 0.37), LVMI (p = 0.063) or LVEDVI + (p = 0.34). The multivariate model of native T1 time including sex and GAA1 + is shown in Table 7. + GAA1 accounted for 7% of the variance in native T1 time. There was one outlier + with a native T1 time of 861 ms (>3 SD below the group mean), but there + was no significant change in the findings when this outlier was excluded from + the regression analyses. GAA1 remained a predictor of native T1 time after + exclusion of diabetic subjects from the analysis (p = 0.025).

\n\n10.1371/journal.pone.0303969.t007\n Multivariate model of native T1 time in adult subjects + with a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variabler value in + univariate analysis\u03B2 in + multivariate modelp value in multivariate modelCumulative adjusted r2
Male sex = 1-0.55-0.420.0010.29
GAA10.480.300.0130.36
\n\n\n

GAA1\u2014number of GAA repeats in the smaller allele + of the FXN gene

\n\n\n

In + children with a normal LVEF, native T1 time was not higher in females (p = + 0.10), GAA1 was a positive correlate of native T1 time (r = 0.45, p = 0.02), + but there were no correlations of native T1 time with any of heart rate (p + = 0.77), AOS (p = 0.75), log SDur (p = 0.79), age (p = 0.10), LVMI (p = 0.89) + or LVEDVI (p = 0.88). GAA1 accounted for 17% of the variance in native T1 + time in children.

\n\n\nPredictors of post + contrast T1 relaxation time in subjects with a normal LVEF\n

In + subjects with a normal LVEF (n = 84), after adjusting for study site, male + sex was a borderline predictor (p = 0.085) of a higher post contrast T1 time, + but there was no difference in post contrast T1 time between children and + adults (p = 0.19). When only adult subjects were analysed, however, both study + site and sex were independent predictors of post contrast T1 time (Table 8). After adjusting for study site + and sex, in adults there were no independent contributions to the prediction + of post contrast T1 time from GAA1 (p = 0.16), AOS (p = 0.59), log SDur (p + = 0.70), heart rate (p = 0.18), age (p = 0.21), LVMI (p = 0.067) or LVEDVI + (p = 0.056). There were no significant predictors of post contrast T1 time + in children (p>0.10 for all).

\n\n10.1371/journal.pone.0303969.t008\n Multivariate model of post contrast T1 time in adult + subjects with a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variabler value in + univariate analysis\u03B2 in + multivariate modelp value in multivariate modelCumulative adjusted r2
Study site0.49-0.50<0.0010.22
Male + sex = 10.330.340.0030.33
\n\n\n\n\nPredictors of extracellular volume fraction in subjects + with a normal LVEF\n

In subjects with a normal LVEF (n = 84), ECV + was higher at the Philadelphia site (p = 0.002), and after adjustment for + study site, it was higher in children than adults (p = 0.013), but not different + between males and females (p = 0.11). In contrast, in separate analysis of + adults with a normal LVEF (n = 57) after adjustment for study site, female + sex was associated with a higher ECV (p = 0.006). In adults after adjustment + for study site and sex, ECV was positively correlated with GAA1 (\u03B2 = + 0.30, p = 0.022), there was a positive correlation with log SDur (\u03B2 = + 0.365, p = 0.009), but only a borderline significant contribution from AOS + (p = 0.069), and no contributions from heart rate (p = 0.25), age (p = 0.21), + LVMI (p = 0.53) or LVEDVI (p = 0.89). A multivariate model of ECV in which + there were independent contributions from study site, sex, GAA1 and log SDur + is shown in Table 9. + GAA1 accounted for 7% of the variance of ECV and log SDur accounted for an + additional 7% of the variance in ECV. When diabetics were excluded from the + analysis, log SDur remained a significant predictor of ECV (p = 0.038), but + GAA1 was no longer significant (p = 0.066). In children there were no significant + predictors of ECV from study site, GAA1, AOS, log SDur, age, LVMI and LVEDVI + (p>0.10 for all).

\n\n10.1371/journal.pone.0303969.t009\n Multivariate model of ECV in adult subjects with + a normal LVEF.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Independent variablesr value in + univariate analysis\u03B2 in + multivariate modelp value in multivariate modelCumulative adjusted r2
Study site0.470. 470.0010.10
Male + sex = 1-0.35-0.240.0500.20
GAA10.440.270.0320.27
Log SDur0.070.340.0130.34
\n\n\n

GAA1\u2014number of GAA repeats in the smaller allele + of the FXN gene; SDur\u2014symptom duration

\n\n\n\n\nReproducibility of T1 measurements\n

There + were 19 randomly selected CMR studies (8 from Melbourne and 11 from Philadelphia, + with gadolinium administered in 18), for which T1 measurements were performed + independently by the investigators at both sites. The results of the two measurements + are shown in Table 10. + There were close correlations of T1 relaxation times for native myocardium + and blood (r = 0.96\u20130.97, n = 19) and for post contrast myocardium and + blood (r = 0.96\u20130.97, n = 18). There were no significant differences + in the repeated measurements for any of these variables.

\n\n10.1371/journal.pone.0303969.t010\n Blinded remeasurements of T1 relaxation times.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
\nNative T1 time (ms)Native + blood time (ms)Post contrast T1 time (ms)Post contrast blood time (ms)
N19191818
Melbourne1017.3\xB139.41583.3\xB1119.8439.7\xB181.3291.3\xB194.0
Philadelphia1014.5\xB147.41574.0\xB1124.3440.4\xB174.3294.8\xB189.6
p value + for comparisons of measurements0.880.500.900.59
r for correlations between measurements0.950.970.960.96
\n\n\n\n\nLate gadolinium enhancement\n

LGE was present + in 19 of the 92 (21%) subjects who had a contrast injection. One of these + 19 only had LGE at an insertion point. The number of segments involved in + the other 18 subjects ranged from 2\u201314 (median of 3). Examples of LGE + in two of these subjects are shown in Figs 1 + and 2. The most common + site for LGE was mid myocardium, but there were also segments with a subepicardial + LGE location. There was only one subject in whom there was evidence of transmural + LGE, this being a subject with a reduced LVEF and LGE in 14 segments. The + walls most frequently involved in the 18 subjects were the anterolateral and + inferolateral walls. Only two subjects had LGE in the septum. LGE was present + in 20% of children and 21% of adults. In subjects with a LVEF \u226555%, there + were 14/83 (17%) with LGE and in subjects with a LVEF<55% there were 5/9 + (56%) with LGE. If more than 4 segments had LGE there was always a reduced + LVEF. In logistic regression analysis of all subjects, a reduced LVEF was + a predictor of LGE presence (p = 0.014), and there were no additional contributions + to the prediction of LGE by sex, age group, GAA1, AOS, log SDur, age, LVEDVI + or LVMI (p>0.10 for all). Similarly, in subjects with a normal LVEF, none + of sex, age group, GAA1, AOS, log SDur, age, LVEDVI or LVMI were predictors + of the presence of LGE (p>0.10 for all).

\n\n10.1371/journal.pone.0303969.g001\n\n\nA cardiac magnetic resonance image showing a + short-axis section of the left ventricle, with the red arrows showing mid + myocardial late gadolinium enhancement in the basal segment of the inferolateral + wall.\n\n\n\n\n10.1371/journal.pone.0303969.g002\n\n\nA cardiac magnetic resonance image showing a + short-axis section of the left ventricle, with the red arrows on the left + showing mid myocardial late gadolinium enhancement in the septal wall and + the red arrows to the right of and underneath the ventricle showing epicardial + late gadolinium enhancement in the anterolateral and inferolateral walls.\n\n\n\n

The 74 subjects without LGE were also + investigated for an association of reduced LVEF with native T1 time and ECV. + A reduced LVEF was not a predictor of native T1 time after adjustment for + sex, heart rate and age group (p>0.50). However, a reduced LVEF remained + a predictor of an increased ECV after adjustment for study site, age, sex + and age group p = 0.04). In adult subjects with a preserved LVEF (n = 48) + and without LGE, after adjustment for sex, GAA1 was still an independent predictor + of native T1 time (\u03B2 = 0.33, p = 0.014) and after adjustment for study + site and log SDur, GAA1 was still an independent predictor of ECV (\u03B2 + = 0.35, p = 0.009).

\n\n\n\nDiscussion\n

Standard + CMR measurements of LVM and LVEDV, in conjunction with T1 mapping techniques + and LGE, were utilised in this study to quantify LV structure and ultrastructure + in subjects with FRDA who were homozygous for GAA expansions in the FXN + gene. The relationships of FRDA genetic severity (GAA1) with macroscopic LV + structure (LVM and LVEDV) and LV ultrastructure (native T1 time, ECV and LGE) + were also investigated. Positive correlations of GAA1 with native T1 relaxation + time and ECV in FRDA subjects with a normal LVEF suggest an association between + the degree of frataxin deficiency and the percentage of diffuse interstitial + LV myocardial fibrous tissue. That these positive correlations were independent + of changes in LVM and LVEDV suggests that T1 mapping techniques provide information + about the LV myocardium which is at least partly independent of the recognised + LV macroscopic structural changes of FRDA. The marker of myocardial replacement + fibrosis, LGE, was found in only 21% of subjects with FRDA and there was substantial + individual variation in the number of LV segments involved. LGE was more frequent + in subjects with a reduced LVEF, but in subjects with a normal LVEF LGE was + not associated with GAA1, LVEDVI or LVMI. The absence of predictors of LGE + suggests that there is a different, and possibly idiosyncratic, mechanism + underlying the development of replacement fibrosis compared to diffuse fibrosis + in the early stages of cardiac disease in FRDA.

\n

In multivariate models + of LVEDV and LVM, BSA was the most important predictor of both variables, + and male sex was an independent predictor of a larger LVM in adults. The contribution + of male sex independently of BSA to a higher LVM is consistent with what is + found in healthy subjects and also with previous CMR and echocardiographic + findings in subjects with FRDA and a normal LVEF [11, 15]. + The findings of the current study again highlight the importance of taking + both body size and sex into account when categorising LV mass in FRDA, something + which has not been performed in some previous studies [4\u20136]. + The finding of an inverse relationship of LVM with age in the present cross-sectional + study is consistent with a previous echocardiographic study [15] and two CMR studies in FRDA [11, 24]. This observation is at present unexplained + and could represent a genuine reduction in LVM during disease progression + [29], survivor + bias, or a combination of both processes. In the present study, in the absence + of age in the model of LVM, log SDur was not a predictor of LVM, this finding + not supporting the hypothesis of a decrease in LVM during disease progression + in FRDA.

\n

In the present study LVM was increased above the normal sex, + age-group and BSA adjusted ranges in 34% of children and 16% of adults. Meyer + et al performed CMR in 41 adults with FRDA, 3 of whom had a LVEF <50%, + and reported an increase in LVMI in 29% of subjects [11]. While this frequency seems substantially + higher in comparison to the adults in the present study, the threshold for + increased LVMI used by Meyer et al was lower, sex differences were not taken + into consideration, and the LVM method used was different to the present study + as it included papillary muscles in the LV volume rather than in the LVM. + Rajagopalan et al also reported an increase in LVMI in FRDA compared to control + subjects in a CMR study which included adults and children with a normal LVEF. + However, the percentage of subjects with a LVMI above the normal range was + not reported in this study, nor was there differentiation made between the + LVMI in males and females [28]. + An increase in LVMI based on echocardiography in subjects with a LVEF >50%, + this study including some subjects from the current study at an earlier time + point, also found higher percentages of age group and sex-adjusted LVMI compared + to the present study (47% of children and 24% of adults) [15]. However, differences in the categorization + of LVMI by CMR and echocardiographic studies is likely to reflect, at least + in part, differences in both accuracy and reproducibility of the two techniques + [39].

\n

More + common than an increase in LVM in the present study was the presence of an + age-group, sex and BSA adjusted LVEDV less than the lower limit of the normal + range, this being found in 39% of subjects (42% of adults and 30% of children). + Thus, a low LVEDV was more common in adults than children, whereas a high + LVM was more common in children than adults. Rajogapalan et al also reported + a lower LVEDV using CMR in FRDA compared to control subjects, but did not + adjust for age group, sex or BSA in their study [28]. + Our finding is consistent with echocardiographic studies which have reported + that the left ventricle is smaller in adults with FRDA and a normal LVEF when + compared to age and sex-matched control subjects [13, 14]. + The current study therefore provides additional evidence that the increases + in wall thickness and LVM in FRDA are associated with a negative LV cavity + remodelling process, as has also been reported in the sarcomeric hypertrophic + cardiomyopathies [40]. + With the combination of an increased LVMI and reduced LVEDVI in our cohort, + an increase in LVMVR was inevitable, and in subjects with a LVEF \u226555%, + the LVMVR was higher than the normal range in 89% of subjects in this study. + Not predictable given the different patterns of increased LVMI and decreased + LVEDVI in adults and children, was that the LVMVR in FRDA was higher in children + than adults.

\n

To the best of our knowledge this is the largest study + of LGE in FRDA, it may be the only study to assess the incidence in LGE in + a group including both adults and children, and it is also the only study + to investigate the predictors of the presence of LGE in FRDA. LGE was only + present in 21% of the subjects in the current study (age range 10\u201349 + years, SDur range up to 36 years), and in contrast to the LVEDVI and LVMI + differences between adults and children, this percentage was similar in adults + and children. LGE was more common in subjects with a reduced LVEF, but it + was not present in all subjects with a reduced LVEF. LGE location was mostly + in the mid myocardium but there was also subepicardial involvement, whereas + transmural LGE involvement was rare. The percentage of subjects with LGE in + previous studies in FRDA has been variable, and both higher and lower percentages + than those of the current study have been reported. Raman et al reported LGE + in 58% of 26 adult subjects with a normal LVEF (age range 18\u201357 years, + SDur not reported) [22]. + Weidemann et al reported LGE in 66% of 32 subjects (age 33\xB113 years, number + of children and SDur not reported), 25% of whom had a LVEF <55% [6]. All those with a + reduced LVEF were positive for LGE, whereas 13/25 (52%) with a normal LVEF + had LGE. That children with FRDA can develop LGE was also previously reported + in 3 children with a normal LVEF by Mavrogeni et al [23]. Takazaki et al reported only 15% LGE + positivity in 27 subjects with a normal LVEF (age 28\xB110 years, number of + children not reported, SDur = 13\xB18 years) [24]. + We can provide no explanation for the substantial variation in LGE positivity + in the different studies, particularly given the absence of important details + about the age groups of the subjects and the SDur for some of the cohorts. + On the other hand, our finding that the mid myocardium was the most common + site for LGE is consistent with the findings of both Raman et al [22] and Weidemann et al [6].

\n

GAA1 is known to be inversely + associated with frataxin levels [25, + 26] and with the + AOS [15], and to + be positively associated with the severity of FRDA cardiac involvement [4, 41]. In the present study, a larger GAA1 + was associated with a larger LVM in adults independently of both sex and body + size. Positive correlations of GAA1 with LVMI have been reported in some [9], but not other [12, 15], echocardiographic studies, and have + not been reported in previous CMR studies [11, + 28]. The discrepancies + between studies regarding the relationship of GAA1 with LVM could be because + the correlation of GAA1 with LVM is relatively weak, that the studies have + been small, and/or that LVMI is less accurately calculated with echocardiography + compared to CMR [42]. + Although LVEDV was smaller than the normal range in 39% of subjects in the + current study, GAA1 was not a predictor of a lower LVEDV. A weak negative + correlation of GAA1 with LVEDV has been reported previously in one echocardiographic + study [15].

\n

We + also examined the relationship of AOS with LV variables in this study, although + issues related to collinearity needed to be considered given the expected + and observed inverse correlation of GAA1 with AOS (r = -0.60). In adults, + LVM was positively correlated with GAA1 and negatively correlated with AOS, + but these associations were not independent. In contrast, in adults, LVEDV + was positively correlated with AOS despite not being correlated with GAA1, + suggesting that the small contribution of AOS to the prediction of LVEDV was + independent of GAA1. The role of SDur was also investigated in this study, + but a longer SDur was only found to be an independent predictor of a smaller + LVEDV in children and a larger ECV in adults.

\n

After adjustment for + study site, the ECV was higher in subjects with a reduced LVEF compared to + those with a normal LVEF, and in the group of subjects with a normal LVEF, + it was higher in children than adults. A recent study reported on ECV in FRDA + subjects using a 3.0T machine [25], + finding it to be 0.36\xB10.05 in 27 subjects with FRDA and a normal LVEF who + were of age 28\xB110 years, this being substantially higher than the value + of 0.25\xB10.02 in the 10 control subjects in that study. In contrast, in + the current study using a 1.5T machine the ECV in subjects with a normal LVEF + was only 0.271\xB10.045 at the Philadelphia site and 0.242\xB10.030 at the + Melbourne site. One possible explanation for the much higher average ECV in + the Takazaki et al study compared to the present study is that it has been + reported that the native T1 time is longer when assessed by a 3T compared + to a 1.5T machine [38]. + Moreover, the type of acquisition sequence has been shown to affect both native + T1 time and ECV [37]. + In addition, the comparison with control subjects in the study of Takazaki + et al was confounded by the small size of the control group (n = 10), with + the imaging performed at a separate site and on a different machine, and the + adequacy of age and sex matching uncertain.

\n

The independent positive + associations of GAA1 with both native T1 time and ECV in the present study + provide evidence supporting a causal relationship between the severity of + frataxin deficiency and an increased ratio of interstitial fibrous/cardiomyocyte + volume in FRDA. However, there were no control subjects in the present study + and there is only limited data in the literature about normal sex and age + values for native T1 and post contrast T1 times and ECV based on the methodology + used in the present study. Our study can therefore provide no definitive evidence + that native T1 time and ECV are increased in FRDA, or how common abnormalities + in native T1 time and ECV might be in FRDA. That sex and age based normal + values are essential for this purpose is supported by studies in healthy subjects + in which males have a lower native T1 time and ECV than females [33\u201338] + and where there is at least a trend, possibly modified by sex, to a lower + native T1 time with older age [33\u201336]. In the present study + the native T1 time was lower in males, and the ECV was lower in adult males + with a normal LVEF, suggesting that similar to healthy subjects, in FRDA there + are not only sex effects on LVMI but also sex effects on the myocardium at + the ultrastructural level.

\n

There is the potential for clinical and + ongoing research roles for T1 mapping in FRDA given that both the native T1 + time and ECV can provide information about the LV myocardium prior to a reduction + of LVEF, and that this information is at least partly independent of the FRDA-associated + LV structural changes. However, the nature, timing and speed of the progression + of changes in T1 mapping variables in FRDA are unknown, and reproducibility + and robustness of T1 mapping techniques for serial testing are uncertain. + In the present study there were differences in post contrast T1 time and ECV + based on the study site even though the machines, protocols and analysis software + were similar at the two sites, and the explanation for this difference remains + uncertain. On the other hand, native T1 time did not appear to be influenced + by the study site, and this simpler technique provides the additional advantages + for the subject in that it does not require gadolinium injection and there + is also less scanning time required.

\n

There are a number of limitations + of this study. The number of children in the study was fewer than the number + of adults and this resulted in lower statistical power to detect effects of + FRDA on CMR variables in children. All studies in FRDA are at least partly + confounded by recruitment being limited to subjects who have presented with + symptoms and also by the absence of those subjects who have had disease progression + and died as a result of the disease. For a CMR study in FRDA there is the + additional limitation that some subjects were ineligible because of the presence + of spinal rods inserted for the treatment of FRDA-associated scoliosis. An + appropriate control group for T1 mapping would have enabled determination + of the frequency of T1 mapping variable abnormalities in FRDA. However, this + was not feasible as it would have required substantial numbers at both sites + to adequately match the age range and sex mix of the FRDA subjects. Moreover, + it would be difficult to justify the use of intravenous gadolinium injection + in normal children, or indeed to even be able to recruit healthy children + for such a study. A further limitation was that the methods for GAA estimation + differed and there is therefore a possibility of systematic differences in + GAA repeat lengths between the different assays. However, this possibility + is more likely to lead to false negative than false positive findings. Moreover, + adjustment for potential differences in techniques between the 2 institutions + has been addressed in part in the regression analyses in which GAA repeat + length was tested an independent variable by the inclusion of the institution + as a dummy variable. Although there were differences in the prediction of + ECV and post GAD T1 time based on institution there were no differences based + on institution in the prediction of native T1 time. This suggests that the + institutional differences could have been related to the post GAD data but + does not support an independent contribution from differences in GAA estimation. + Lastly, there is uncertainty regarding the mechanistic significance of some + of the regression analysis findings due to collinearity between GAA1 and AOS + and also between age and SDur. That there were independent effects of age + from SDur and vice versa was suggested by the evidence that age, but not SDur, + was an independent predictor for LVM in adult subjects with a normal LVEF + and that SDur, but not age, was an independent predictor of ECV in adult subjects + with a normal LVEF.

\n

In conclusion, there is an association between + diffuse interstitial LV myocardial fibrosis and genetic severity in FRDA, + with this effect being independent of the FRDA-associated LV changes in LVEDVI + and LVMI. Localised replacement fibrosis was found in a minority of subjects + with FRDA, this minority including both children and adults, and subjects + with and without a reduction of LVEF. In contrast to T1 mapping variables, + LGE in subjects with a normal LVEF was not associated with genetic severity, + and was also not associated with LVM or LVEDV, consistent with the development + of LGE in FRDA having an idiosyncratic element. There is the potential for + roles of both T1 mapping and LGE in the assessment and monitoring of the cardiomyopathy + of FRDA, but information will be required about their prognostic significance, + and more data will also be required regarding the reproducibility of T1 mapping + variables.

\n\n\nSupporting + information\n\n\n\nAnthropomorphic + and cardiac magnetic resonance data in subjects with Friedreich ataxia.\n

(XLSX)

\n\n\n\n\n\n\n

We + would like to thank all subjects who generously volunteered their time to + participate in this study. In addition, we would like to thank Ms Genevieve + Tai from the Murdoch Children\u2019s Research Institute who provided laboratory + and research support and Michael Kean and the staff of the Royal Children\u2019s + Hospital MRI department for their expertise and support of this study.

\n\n\nAbbreviations\n\nAOS\n

age + at onset of symptoms

\n\nBSA\n

body + surface area

\n\nCMR\n

cardiac + magnetic resonance

\n\nECV\n

left + ventricular myocardial extracellular volume fraction

\n\nFRDA\n

Friedreich + ataxia

\n\nGAA1\n

number + of GAA repeats in the smaller allele of the FXN gene

\n\nLGE\n

late + gadolinium enhancement

\n\nLV\n

left + ventricular

\n\nLVEDV\n

left + ventricular end-diastolic volume

\n\nLVEDVI\n

left + ventricular end-diastolic volume indexed to body surface area

\n\nLVEF\n

left + ventricular ejection fraction

\n\nLVESV\n

left + ventricular end-systolic volume

\n\nLVM\n

left + ventricular mass

\n\nLVMI\n

left + ventricular mass indexed to body surface area

\n\nSDur\n

symptom + duration

\n\nSV\n

stroke + volume

\n\n\n\n\nReferences\nDelatycki + MB, Corben + LA. Clinical features of + Friedreich ataxia. J Child Neurol. 2012;27(9):1133\u20137. + doi: 10.1177/0883073812448230 22752493\nHewer + RL. Study of fatal cases + of Friedreich\u2019s ataxia. British Medical Journal. + 1968;3(5619):649\u201352. + doi: 10.1136/bmj.3.5619.649 5673214\nTsou + AY, Paulsen + EK, Lagedrost + SJ, Perlman + SL, Mathews + KD, Wilmot + GR, et al. Mortality + in Friedreich ataxia. J Neurol Sci. 2011;307(1\u20132):46\u20139. + doi: 10.1016/j.jns.2011.05.023 21652007\nPousset + F, Legrand + L, Monin + ML, Ewenczyk + C, Charles + P, Komajda + M, et al. A + 22-Year Follow-up Study of Long-term Cardiac Outcome and Predictors of Survival + in Friedreich Ataxia. JAMA Neurol. 2015:1\u20138. + doi: 10.1001/jamaneurol.2015.1855 26414159\nWeidemann + F, Rummey + C, Bijnens + B, Stork + S, Jasaityte + R, Dhooge + J, et al. The + heart in Friedreich ataxia: Definition of cardiomyopathy, disease severity, + and correlation with neurological symptoms. Circulation. + 2012;125:1626\u201334. + doi: 10.1161/CIRCULATIONAHA.111.059477 + 22379112\nWeidemann + F, Liu + D, Hu + K, Florescu + C, Niemann + M, Herrmann + S, et al. The + cardiomyopathy in Friedreich\u2019s ataxia\u2014New biomarker for staging + cardiac involvement. Int J Cardiol. 2015;194:50\u20137. + doi: 10.1016/j.ijcard.2015.05.074 26005806\nLegrand + L, Diallo + A, Monin + ML, Ewenczyk + C, Charles + P, Isnard + R, et al. Predictors + of left ventricular dysfunction in Friedreich\u2019s Ataxia in a 16-year observational + study. Am J Cardiovasc Drugs. 2020;20(2):209\u201316. + doi: 10.1007/s40256-019-00375-z 31650522\nHewer + R. The heart in Friedreich\u2019s + ataxia. Br Heart J. 1969;31(1):5\u201314. + doi: 10.1136/hrt.31.1.5 5774037\nIsnard + R, Kalotka + H, Durr + A, Cossee + M, Schmitt + M, Pousset + F, et al. Correlation + between left ventricular hypertrophy and GAA trinucleotide repeat length in + Friedreichs ataxia. Circulation. 1997;95(9):2247\u20139. + doi: 10.1161/01.cir.95.9.2247 9142000\nDutka + DP, Donnelly + JE, Nihoyannopoulos + P, Oakley + CM, Nunez + DJ. Marked variation in the + cardiomyopathy associated with Friedreich\u2019s ataxia. Heart + (British Cardiac Society). 1999;81(2):141\u20137. + doi: 10.1136/hrt.81.2.141 9922348\nMeyer + C, Schmid + G, Gorlitz + S, Ernst + M, Wilkens + C, Wilhelms + I, et al. Cardiomyopathy + in Friedreich\u2019s ataxia-assessment by cardiac MRI. Mov + Disord. 2007;22(11):1615\u201322. + doi: 10.1002/mds.21590 17546670\nRegner + SR, Lagedrost + SJ, Plappert + T, Paulsen + EK, Friedman + LS, Snyder + ML, et al. Analysis + of echocardiograms in a large heterogeneous cohort of patients with Friedreich + ataxia. Am J Cardiol. 2012;109(3):401\u20135. + doi: 10.1016/j.amjcard.2011.09.025 22078220\nMottram + PM, Delatycki + MB, Donelan + L, Gelman + JS, Corben + L, Peverill + RE. Early changes in left + ventricular long-axis function in Friedreich ataxia: relation with the FXN + gene mutation and cardiac structural change. J Am + Soc Echocardiogr. 2011;24(7):782\u20139. + doi: 10.1016/j.echo.2011.04.004 21570254\nDedobbeleer + C, Rai + M, Donal + E, Pandolfo + M, Unger + P. Normal left ventricular + ejection fraction and mass but subclinical myocardial dysfunction in patients + with Friedreich\u2019s ataxia. Eur Heart J Cardiovasc + Imaging. 2012;13(4):346\u201352. + doi: 10.1093/ejechocard/jer267 22127629\nPeverill + RE, Romanelli + G, Donelan + L, Hassam + R, Corben + LA, Delatycki + MB. Left ventricular structural + and functional changes in Friedreich ataxia\u2014Relationship with body size, + sex, age and genetic severity. PLoS ONE. + 2019;14(11):e0225147. + doi: 10.1371/journal.pone.0225147 31721791\nMewton + N, Liu + CY, Croisille + P, Bluemke + D, Lima + JA. Assessment of myocardial + fibrosis with cardiovascular magnetic resonance. J + Am Coll Cardiol. 2011;57(8):891\u2013903. + doi: 10.1016/j.jacc.2010.11.013 21329834\nWhite + SK, Sado + DM, Flett + AS, Moon + JC. Characterising the myocardial + interstitial space: the clinical relevance of non-invasive imaging. + Heart. 2012;98(10):773\u20139. + doi: 10.1136/heartjnl-2011-301515 22422587\nFlett + AS, Hayward + MP, Ashworth + MT, Hansen + MS, Taylor + AM, Elliott + PM, et al. Equilibrium + contrast cardiovascular magnetic resonance for the measurement of diffuse + myocardial fibrosis: preliminary validation in humans. Circulation. + 2010;122(2):138\u201344. + doi: 10.1161/CIRCULATIONAHA.109.930636 + 20585010\nIles + L, Pfluger + H, Phrommintikul + A, Cherayath + J, Aksit + P, Gupta + SN, et al. Evaluation + of diffuse myocardial fibrosis in heart failure with cardiac magnetic resonance + contrast-enhanced T1 mapping. J Am Coll Cardiol. + 2008;52(19):1574\u201380. + doi: 10.1016/j.jacc.2008.06.049 19007595\nJerosch-Herold + M, Sheridan + DC, Kushner + JD, Nauman + D, Burgess + D, Dutton + D, et al. Cardiac + magnetic resonance imaging of myocardial contrast uptake and blood flow in + patients affected with idiopathic or familial dilated cardiomyopathy. + Am J Physiol Heart Circ Physiol. 2008;295(3):H1234\u2013H42. + doi: 10.1152/ajpheart.00429.2008 18660445\nUgander + M, Oki + AJ, Hsu + LY, Kellman + P, Greiser + A, Aletras + AH, et al. Extracellular + volume imaging by magnetic resonance imaging provides insights into overt + and sub-clinical myocardial pathology. Eur Heart J. + 2012;33(10):1268\u201378. + doi: 10.1093/eurheartj/ehr481 22279111\nRaman + SV, Phatak + K, Hoyle + JC, Pennell + ML, McCarthy + B, Tran + T, et al. Impaired + myocardial perfusion reserve and fibrosis in Friedreich ataxia: a mitochondrial + cardiomyopathy with metabolic syndrome. Eur Heart + J. 2011;32(5):561\u20137. + doi: 10.1093/eurheartj/ehq443 21156720\nMavrogeni + S, Giannakopoulou + A, Katsalouli + M, Pons + RM, Papavasiliou + A, Kolovou + G, et al. Friedreich\u2019s + ataxia: Case series and the additive value of cardiovascular magnetic resonance. + J Neuromuscul Dis. 2020;7(1):61\u20137. + doi: 10.3233/JND-180373 31796683\nTakazaki + KAG, Quinaglia + T, Venancio + TD, Martinez + ARM, Shah + RV, Neilan + TG, et al. Pre-clinical + left ventricular myocardial remodeling in patients with Friedreich\u2019s + ataxia: A cardiac MRI study. PLoS ONE. 2021;16(3):e0246633. + doi: 10.1371/journal.pone.0246633 33770103\nCampuzano + V, Montermini + L, Lutz + Y, Cova + L, Hindelang + C, Jiralerspong + S, et al. Frataxin + is reduced in Friedreich ataxia patients and is associated with mitochondrial + membranes. Hum Mol Genet. 1997;6(11):1771\u201380. + doi: 10.1093/hmg/6.11.1771 9302253\nDeutsch + EC, Santani + AB, Perlman + SL, Farmer + JM, Stolle + CA, Marusich + MF, et al. A + rapid, noninvasive immunoassay for frataxin: utility in assessment of Friedreich + ataxia. Mol Genet Metab. 2010;101(2\u20133):238\u201345. + doi: 10.1016/j.ymgme.2010.07.001 20675166\nBit-Avragim + N, Perrot + A, Schols + L, Hardt + C, Kreuz + FR, Zuhlke + C, et al. The + GAA repeat expansion in intron 1 of the frataxin gene is related to the severity + of cardiac manifestation in patients with Friedreich\u2019s ataxia. + J Mol Med. 2001;78(11):626\u201332. + doi: 10.1007/s001090000162 11269509\nRajagopalan + B, Francis + JM, Cooke + F, Korlipara + LV, Blamire + AM, Schapira + AH, et al. Analysis + of the factors influencing the cardiac phenotype in Friedreich\u2019s ataxia. + Mov Disord. 2010;25(7):846\u201352. + doi: 10.1002/mds.22864 20461801\nRibai + P, Pousset + F, Tanguy + ML, Rivaud-Pechoux + S, Le + B, Gasparini + F, et al. Neurological, + cardiological, and oculomotor progression in 104 patients with Friedreich + ataxia during long-term follow-up. Arch Neurol. + 2007;64(4):558\u201364. + doi: 10.1001/archneur.64.4.558 17420319\nKawel-Boehm + N, Hetzel + SJ, Ambale-Venkatesh + B, Captur + G, Francois + CJ, Jerosch-Herold + M, et al. Reference + ranges (\"normal values\") for cardiovascular magnetic resonance (CMR) in + adults and children: 2020 update. J Cardiovasc Magn + Reson. 2020;22(1):87. + doi: 10.1186/s12968-020-00683-3 33308262\nTreibel + TA, Fridman + Y, Bering + P, Sayeed + A, Maanja + M, Frojdh + F, et al. Extracellular + Volume Associates With Outcomes More Strongly Than Native or Post-Contrast + Myocardial T1. JACC Cardiovasc Imaging. 2020;13(1 + Pt 1):44\u201354.\nTreibel + TA, Fontana + M, Maestrini + V, Castelletti + S, Rosmini + S, Simpson + J, et al. Automatic + Measurement of the Myocardial Interstitium: Synthetic Extracellular Volume + Quantification Without Hematocrit Sampling. JACC Cardiovasc + Imaging. 2016;9(1):54\u201363. + doi: 10.1016/j.jcmg.2015.11.008 26762875\nLiu + CY, Liu + YC, Wu + C, Armstrong + A, Volpe + GJ, Van + der Geest RJ, et al. + Evaluation of age-related interstitial myocardial fibrosis + with cardiac magnetic resonance contrast-enhanced T1 mapping: MESA (Multi-Ethnic + Study of Atherosclerosis). J Am Coll Cardiol. + 2013;62(14):1280\u20137. + doi: 10.1016/j.jacc.2013.05.078 23871886\nPiechnik + SK, Ferreira + VM, Lewandowski + AJ, Ntusi + NA, Banerjee + R, Holloway + C, et al. Normal + variation of magnetic resonance T1 relaxation times in the human population + at 1.5 T using ShMOLLI. J Cardiovasc Magn Reson. + 2013;15:13. doi: 10.1186/1532-429X-15-13 23331520\nRauhalammi + SM, Mangion + K, Barrientos + PH, Carrick + DJ, Clerfond + G, McClure + J, et al. Native + myocardial longitudinal (T1) relaxation time: Regional, age, and sex associations + in the healthy adult heart. J Magn Reson Imaging. + 2016;44(3):541\u20138. + doi: 10.1002/jmri.25217 26946323\nRoy + C, Slimani + A, de + Meester C, Amzulescu + M, Pasquet + A, Vancraeynest + D, et al. Age + and sex corrected normal reference values of T1, T2 T2* and ECV in healthy + subjects at 3T CMR. J Cardiovasc Magn Reson. + 2017;19(1):72.\nRosmini + S, Bulluck + H, Captur + G, Treibel + TA, Abdel-Gadir + A, Bhuva + AN, et al. Myocardial + native T1 and extracellular volume with healthy ageing and gender. + Eur Heart J Cardiovasc Imaging. 2018;19(6):615\u201321. + doi: 10.1093/ehjci/jey034 29617988\nGranitz + M, Motloch + LJ, Granitz + C, Meissnitzer + M, Hitzl + W, Hergan + K, et al. Comparison + of native myocardial T1 and T2 mapping at 1.5T and 3T in healthy volunteers: + Reference values and clinical implications. Wien Klin + Wochenschr. 2019;131(7\u20138):143\u201355.\nMyerson + SG, Bellenger + NG, Pennell + DJ. Assessment of left ventricular + mass by cardiovascular magnetic resonance. Hypertension. + 2002;39(3):750\u20135. + doi: 10.1161/hy0302.104674 11897757\nHaland + TF, Hasselberg + NE, Almaas + VM, Dejgaard + LA, Saberniak + J, Leren + IS, et al. The + systolic paradox in hypertrophic cardiomyopathy. Open + Heart. 2017;4(1):e000571. + doi: 10.1136/openhrt-2016-000571 28674623\nPeverill + RE, Donelan + L, Corben + LA, Delatycki + MB. Differences in the determinants + of right ventricular and regional left ventricular long-axis dysfunction in + Friedreich ataxia. PLoS ONE. 2018;13(12):e0209410. + doi: 10.1371/journal.pone.0209410 30596685\nBellenger + NG, Davies + LC, Francis + JM, Coats + AJ, Pennell + DJ. Reduction in sample size + for studies of remodeling in heart failure by the use of cardiovascular magnetic + resonance. J Cardiovasc Magn Reson. 2000;2(4):271\u20138. + doi: 10.3109/10976640009148691 11545126\n\n\n
" + headers: + Accept-Ranges: + - bytes + Cache-Control: + - private, max-age=0 + Content-Disposition: + - attachment; filename=journal.pone.0303969.xml + Content-Length: + - '137547' + Content-Type: + - application/xml + Date: + - Wed, 23 Sep 2026 16:31:43 GMT + ETag: + - '"dc290c36e2a2202fc3d5570aef82f5a8"' + Expires: + - Wed, 23 Sep 2026 16:31:43 GMT + Last-Modified: + - Thu, 30 May 2024 17:27:22 GMT + Server: + - UploadServer + X-GUploader-UploadID: + - AP6rU80XNNG3HEfoKdxg0ymk5ArgvnEvcIv8ByoFe2WMGctmHqpsZ0iBCw_SFDV6U09AWASwVkX8rc0 + x-goog-generation: + - '1717090042542921' + x-goog-hash: + - crc32c=YWEAdQ== + - md5=3CkMNuKiIC/D1VcK74L1qA== + x-goog-metageneration: + - '1' + x-goog-storage-class: + - STANDARD + x-goog-stored-content-encoding: + - identity + x-goog-stored-content-length: + - '137547' + status: + code: 200 + message: OK version: 1 diff --git a/ace/tests/test_ace.py b/ace/tests/test_ace.py index 6d62e96..d18a613 100644 --- a/ace/tests/test_ace.py +++ b/ace/tests/test_ace.py @@ -183,27 +183,40 @@ def test_database_processing_stream(db, test_data_path): @pytest.mark.vcr(record_mode="once") -def test_journal_scraping(test_data_path): +def test_journal_scraping(test_data_path, source_manager): scrape_path = join(test_data_path, 'scrape_test') - os.makedirs(scrape_path, exist_ok=True) + # Start clean: index_pmids makes the scraper skip anything already on disk, + # so articles left behind by an earlier failed run would send it past the + # PMIDs the cassette covers. + shutil.rmtree(scrape_path, ignore_errors=True) + os.makedirs(scrape_path) # Test with PLoS ONE because it's OA scraper = scrape.Scraper(scrape_path) - scraper.retrieve_articles( - 'PLoS ONE', - delay=5.0, - mode='requests', - search='fmri', - limit=2, - index_pmids=True, - skip_pubmed_central=False, - invalid_article_log_file=join(scrape_path, 'invalid_articles.log'), - prefer_pmc_source=True, - ) - # For now just check to make sure we have expected number of files in the directory - plos_dir = join(scrape_path, 'html/PLoS ONE/') - n_files = len([name for name in os.listdir(plos_dir) if os.path.isfile(plos_dir + name)]) - shutil.rmtree(scrape_path) - assert n_files == 2 + try: + scraper.retrieve_articles( + 'PLoS ONE', + delay=5.0, + mode='requests', + search='fmri', + limit=2, + index_pmids=True, + skip_pubmed_central=False, + invalid_article_log_file=join(scrape_path, 'invalid_articles.log'), + prefer_pmc_source=True, + ) + plos_dir = join(scrape_path, 'html/PLoS ONE/') + names = [name for name in os.listdir(plos_dir) if os.path.isfile(plos_dir + name)] + docs = [open(join(plos_dir, name)).read() for name in names] + finally: + shutil.rmtree(scrape_path, ignore_errors=True) + + assert len(names) == 2 + # Each file must be the PLoS NLM XML, not the PubMed abstract page we get + # bounced to when E-utilities has no LinkOut record for an article. + for name, doc in zip(names, docs): + assert '", "PLoS ONE") == landing + + +def test_validate_scrape_flags_contentless_recaptcha_page(): + html = ( + 'Just a moment' + '
' + '' + ) + assert scrape._validate_scrape(html) is False + + @pytest.mark.parametrize( "pmid,expected_source", [ diff --git a/ace/tests/weird_data/38814901_pubmed_recaptcha_widget.html b/ace/tests/weird_data/38814901_pubmed_recaptcha_widget.html new file mode 100644 index 0000000..29b4c65 --- /dev/null +++ b/ace/tests/weird_data/38814901_pubmed_recaptcha_widget.html @@ -0,0 +1,3655 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + Insights into the effects of Friedreich ataxia on the left ventricle using T1 mapping and late gadolinium enhancement - PubMed + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + Skip to main page content + + + +
+
+
+
+ U.S. flag +

An official website of the United States government

+ +
+
+
+
+ Dot gov +
+

+ The .gov means it’s official. +
+ Federal government websites often end in .gov or .mil. Before + sharing sensitive information, make sure you’re on a federal + government site. +

+
+
+
+ Https +
+

+ The site is secure. +
+ The https:// ensures that you are connecting to the + official website and that any information you provide is encrypted + and transmitted securely. +

+
+
+
+
+
+
+ +
+Access keys +NCBI Homepage +MyNCBI Homepage +Main Content +Main Navigation +
+
+
+
+ + + + + + + + + + + + + +
+
+
+
+
+ + +
+ +
+
+
+ + + +
+ + + +
+ +
+
+
+ +
+ + + + + + + + + + + + + + + + + + + + + + + +
+
+ +
+ + +
+ + + + + + + + + + + + + + + + + + +
+ + +
+ +
+ + + + +
+ +
. 2024 May 30;19(5):e0303969. + +
+ + + + + doi: 10.1371/journal.pone.0303969. + + + + + + eCollection 2024. + + + + + +
+ + +

+ + + + + + + Insights into the effects of Friedreich ataxia on the left ventricle using T1 mapping and late gadolinium enhancement + + + + +

+ + + +
+ + + + + +
+ + + +
+ Roger E Peverill  + 1 + , Kimberly Y Lin  + 2 + , Mark A Fogel  + 2 + , Michael M H Cheung  + 3 +   + 4 +   + 5 + , W Stuart Moir  + 1 + , Louise A Corben  + 6 + , Glenn Cahoon  + 3 + , Martin B Delatycki  + 6 +   + 7 + +
+ + +
+ + + + + + + + +
+ + + +
+ + + + Affiliations + + + + + +
+ + + + + + + + + + + + +
+
+ + + + +

+ + + + + + + Insights into the effects of Friedreich ataxia on the left ventricle using T1 mapping and late gadolinium enhancement + + + + +

+ + + +
+ + + + Roger E Peverill et al. + + + + + + + PLoS One. + + + + . + + + + +
+ + +
+ + + + +
+ + + +
+ +
+ +
+ + +
+ + + +
+ + + + + + + +
+ +

+ Abstract + +

+ + + +
+ + + + + +

+ + + Background: + + + The left ventricular (LV) changes which occur in Friedreich ataxia (FRDA) are incompletely understood. +

+ + + + + + + + + + +

+ + + Methods: + + + Cardiac magnetic resonance (CMR) imaging was performed using a 1.5T scanner in subjects with FRDA who are homozygous for an expansion of an intron 1 GAA repeat in the FXN gene. Standard measurements were performed of LV mass (LVM), LV end-diastolic volume (LVEDV) and LV ejection fraction (LVEF). Native T1 relaxation time and the extracellular volume fraction (ECV) were utilised as markers of left ventricular (LV) diffuse myocardial fibrosis and late gadolinium enhancement (LGE) was utilised as a marker of LV replacement fibrosis. FRDA genetic severity was assessed using the shorter FXN GAA repeat length (GAA1). +

+ + + + + + + + + + +

+ + + Results: + + + There were 93 subjects with FRDA (63 adults, 30 children, 54% males), 9 of whom had a reduced LVEF (<55%). A LVEDV below the normal range was present in 39%, a LVM above the normal range in 22%, and an increased LVM/LVEDV ratio in 89% subjects. In adults with a normal LVEF, there was an independent positive correlation of LVM with GAA1, and a negative correlation with age, but no similar relationships were seen in children. GAA1 was positively correlated with native T1 time in both adults and children, and with ECV in adults, all these associations independent of LVM and LVEDV. LGE was present in 21% of subjects, including both adults and children, and subjects with and without a reduced LVEF. None of GAA1, LVM or LVEDV were predictors of LGE. +

+ + + + + + + + + + +

+ + + Conclusion: + + + An association between diffuse interstitial LV myocardial fibrosis and genetic severity in FRDA was present independently of FRDA-related LV structural changes. Localised replacement fibrosis was found in a minority of subjects with FRDA and was not associated with LV structural change or FRDA genetic severity in subjects with a normal LVEF. +

+ + + + + + +
+ + + + + + + + + + + +
+ + + + + + + + + +

+ PubMed Disclaimer +

+ +
+

+ Conflict of interest statement +

+ +
+ + + +

The authors have declared that no competing interests exist.

+ + +
+
+ + + + + + + + + + + + + + + + + + +
+

+ MeSH terms +

+ + +
+ + + + + + + + + + + + + + + + + + + + + + + + +
+ + + + + + + +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + \ No newline at end of file