PNAS PLUS

Persistent cannabis users show neuropsychological

decline from childhood to midlife

SEE COMMENTARY
Madeline H. Meiera,b,1, Avshalom Caspia,b,c,d,e, Antony Amblere,f, HonaLee Harringtonb,c,d, Renate Houtsb,c,d,
Richard S. E. Keefed, Kay McDonaldf, Aimee Wardf, Richie Poultonf, and Terrie E. Moffitta,b,c,d,e
a
Duke Transdisciplinary Prevention Research Center, Center for Child and Family Policy, bDepartment of Psychology and Neuroscience, and cInstitute for
Genome Sciences and Policy, Duke University, Durham, NC 27708; dDepartment of Psychiatry and Behavioral Sciences, Duke University Medical Center,
Durham, NC 27710; eSocial, Genetic, and Developmental Psychiatry Centre, Institute of Psychiatry, King’s College London, London SE5 8AF, United Kingdom;
and fDunedin Multidisciplinary Health and Development Research Unit, Department of Preventive and Social Medicine, School of Medicine, University of
Otago, Dunedin 9054, New Zealand

Edited by Michael I. Posner, University of Oregon, Eugene, OR, and approved July 30, 2012 (received for review April 23, 2012)

Recent reports show that fewer adolescents believe that regular nence from cannabis. There are two commonly cited potential
cannabis use is harmful to health. Concomitantly, adolescents are limitations of this approach. One is the absence of data on
initiating cannabis use at younger ages, and more adolescents are initial, precannabis-use neuropsychological functioning. It is
using cannabis on a daily basis. The purpose of the present study possible that differences in test performance between cannabis
was to test the association between persistent cannabis use and users and controls are attributable to premorbid rather than
neuropsychological decline and determine whether decline is cannabis-induced deficits (17–20). A second limitation is re-
concentrated among adolescent-onset cannabis users. Participants liance on retrospectively reported quantity, frequency, dura-
were members of the Dunedin Study, a prospective study of tion, and age-of-onset of cannabis use, often inquired about
a birth cohort of 1,037 individuals followed from birth (1972/1973) years after initiation of heavy use.
to age 38 y. Cannabis use was ascertained in interviews at ages A prospective, longitudinal investigation of the association
18, 21, 26, 32, and 38 y. Neuropsychological testing was conducted between cannabis use and neuropsychological impairment could
at age 13 y, before initiation of cannabis use, and again at age redress these limitations and strengthen the existing evidence
38 y, after a pattern of persistent cannabis use had developed. base by assessing neuropsychological functioning in a sample of
Persistent cannabis use was associated with neuropsychological youngsters before the onset of cannabis use, obtaining pro-
decline broadly across domains of functioning, even after control- spective data on cannabis use as the sample is followed over
ling for years of education. Informants also reported noticing more a number of years, and readministering neuropsychological tests

PSYCHOLOGICAL AND
COGNITIVE SCIENCES
cognitive problems for persistent cannabis users. Impairment was after some members of the sample have developed a pattern of
concentrated among adolescent-onset cannabis users, with more long-term cannabis use. To our knowledge, only one prospective,
persistent use associated with greater decline. Further, cessation longitudinal study of the effects of cannabis on neuropsychol-
of cannabis use did not fully restore neuropsychological function- ogical functioning has been conducted (21), and, in this study,
ing among adolescent-onset cannabis users. Findings are sugges- the sample was small and the average duration of regular can-
tive of a neurotoxic effect of cannabis on the adolescent brain nabis use was only 2 y.
and highlight the importance of prevention and policy efforts In the present study, we investigated the association between
targeting adolescents. persistent cannabis use—prospectively assessed over 20 y—and
neuropsychological functioning in a birth cohort of 1,037 indi-
marijuana | longitudinal | cognition viduals. Study members underwent neuropsychological testing in
1985 and 1986 before the onset of cannabis use and again in
2010–2012, after some had developed a persistent pattern of
C annabis, the most widely used illicit drug in the world, is
increasingly being recognized for both its toxic and its ther-
apeutic properties (1). Research on the harmful and beneficial
cannabis use. We tested six hypotheses. First, we tested the
“cognitive decline” hypothesis that persistent cannabis users
effects of cannabis use is important because it can inform deci- evidence greater decline in test performance from childhood to
sions regarding the medicinal use and legalization of cannabis, adulthood than nonusers. By examining within-person change in
and the results of these decisions will have major public-health neuropsychological functioning, any effect of premorbid deficits
consequences. As debate surrounding these issues continues in on later (postcannabis-initiation) test performance was nullified.
the United States and abroad, new findings concerning the harmful Second, we tested the “specificity” hypothesis to address whether
effects of cannabis on neuropsychological functioning are emerging. impairment is confined to specific neuropsychological domains
Accumulating evidence suggests that long-term, heavy can- or whether it is more global. To test this hypothesis, we admin-
nabis use may cause enduring neuropsychological impairment— istered multiple tests for each of five specific domains, as dif-
impairment that persists beyond the period of acute intoxication ferent tests may be differentially sensitive to cannabis-associated
(2). Studies of long-term, heavy cannabis users fairly consistently neuropsychological impairment. In conducting our analyses, we
show that these individuals perform worse on neuropsychological tested alternative explanations for the association between per-
tests (2–5), and some (6–8) but not all (9) studies suggest that
impairment may remain even after extended periods of absti-
nence. The magnitude and persistence of impairment may de- Author contributions: M.H.M., A.C., and T.E.M. designed research; M.H.M., A.C., A.A.,
pend on factors such as the quantity, frequency, duration, and H.H., R.H., R.S.E.K., K.M., A.W., R.P., and T.E.M. performed research; M.H.M., A.C., R.H.,
and T.E.M. analyzed data; and M.H.M., A.C., and T.E.M. wrote the paper.
age-of-onset of cannabis use (2), as more severe and enduring
impairment is evident among individuals with more frequent and The authors declare no conflict of interest.

prolonged heavy use and a younger age-of-onset (3, 6, 8, 10–16). This article is a PNAS Direct Submission.
The extant evidence base draws on case–control studies of See Commentary on page 15970.
recruited cannabis users and comparison subjects. These stud- 1
To whom correspondence should be addressed. E-mail: [email protected].
ies screen participants for potential confounding factors, See Author Summary on page 15980 (volume 109, number 40).
such as alcohol and drug dependence, and compare them on This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
neuropsychological test performance after a period of absti- 1073/pnas.1206820109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1206820109 PNAS | Published online August 27, 2012 | E2657–E2664

sistent cannabis use and neuropsychological functioning by ruling IQ decline was most pronounced among the most persistent
out potential confounding effects of (i) acute or residual can- cannabis-dependence group (i.e., the 3+ group; n = 38), but the
nabis intoxication, (ii) tobacco dependence, (iii) hard-drug de- effect of persistent cannabis dependence on IQ decline was not
pendence (e.g., heroin, cocaine, amphetamines), (iv) alcohol solely attributable to this group. For example, the association
dependence, and (v) schizophrenia. Third, we tested the “edu- between persistent cannabis dependence and full-scale IQ de-
cation” hypothesis that persistent cannabis users experience cline was still apparent after excluding the study members with
neuropsychological decline simply because they have eschewed 3+ cannabis-dependence diagnoses from the analysis (t = −2.94,
academics and other opportunities for learning. Recent evidence P = 0.0034). Table S1 shows parallel results for persistent reg-
suggests that staying in school can boost one’s intelligence quo- ular cannabis use and persistent cannabis dependence.
tient (IQ) (22), and cannabis users tend to receive less schooling
than nonusers (23). Therefore, we tested whether the association Is Impairment Specific to Certain Neuropsychological Domains or Is It
between persistent cannabis use and neuropsychological decline Global? Table 3 shows the effects of persistent cannabis de-
remained after controlling for years of education. Fourth, we pendence on five different areas of mental function assessed at
queried third-party informants to test the “everyday cognition” age 38 y. Effects represent mean neuropsychological test per-
hypothesis that cannabis-induced neuropsychological impair- formance at age 38 y, adjusted for childhood IQ. Across
ment translates into functional problems in daily life. Fifth, we different areas of mental function, study members with more
tested the “developmental vulnerability” hypothesis that indi- persistent cannabis dependence generally showed greater
viduals who begin cannabis use as adolescents are particularly neuropsychological impairment. Inspection of the means sug-
vulnerable to the effects of persistent cannabis use on neuro- gests that the greatest impairments were for the domains of
psychological functioning, as evidence suggests that cannabis has executive functioning and processing speed. To test whether
especially toxic effects on the developing brain (24–31). Sixth, we impairment was relatively greater for certain domains, we
tested the “recovery” hypothesis that former persistent users who compared cannabis-associated neuropsychological impairment
quit or reduce their cannabis use may be able to restore their across the four Wechsler Adult Intelligence Scale-IV (WAIS-
neuropsychological health. IV) indexes (i.e., working memory index, processing speed in-
dex, perceptual reasoning index, and verbal comprehension
Results index), which share psychometric properties (i.e., reliability)
Do Study Members with More Persistent Cannabis Use Show Greater important for such a test. Using a model-fitting approach, we
IQ Decline? Table 1 (far right column) shows effect sizes for within- fitted (i) a model allowing the association between persistent
person IQ change from childhood to adulthood as a function of cannabis dependence and age-38 neuropsychological impair-
persistent cannabis dependence. In this analysis, each study ment, adjusted for childhood IQ and sex, to vary across the four
member served as his or her own control; given that the groups WAIS-IV indexes and (ii) a model equating this association
were not equivalent on childhood IQ, we accounted for premorbid across the four WAIS-IV indexes. Results showed that associ-
IQ differences by looking at IQ change from childhood to age 38 y. ations between persistent cannabis dependence and all four
Study members with more persistent cannabis dependence WAIS-IV indexes could be equated without a resultant de-
showed greater IQ decline. For example, study members who terioration in model fit (Δχ2 = 2.13, df = 3, P = 0.55), which
never used cannabis experienced a slight increase in IQ, whereas suggests that impairment was not statistically significantly dif-
those who diagnosed with cannabis dependence at one, two, or ferent across neuropsychological domains.
three or more study waves experienced IQ declines of −0.11,
−0.17, and −0.38 SD units, respectively. An IQ decline of −0.38 SD Is Impairment Attributable to Persistent Cannabis Use or Are There
units corresponds to a loss of ∼6 IQ points, from 99.68 to 93.93. Alternative Explanations? We ruled out six alternative explan-
Results of analyses for persistent cannabis dependence and per- ations for the observed effects of persistent cannabis use on
sistent regular cannabis use were similar (Table 1). neuropsychological functioning, namely that these effects could
Table 2 expands the analysis by showing results for the sub- be explained by (i) past 24-h cannabis use, (ii) past-week can-
tests of different cognitive abilities that constitute the IQ. Per- nabis use, (iii) persistent tobacco dependence, (iv) persistent
sistent cannabis dependence was associated with greater decline hard-drug dependence, (v) persistent alcohol dependence, and
on the majority of the subtests. (vi) schizophrenia. We recalculated the mean change in full-

Table 1. IQ before and after cannabis use

N % male Age 7–13 full-scale IQ Age 38 full-scale IQ Δ IQ effect size*

Persistence of cannabis dependence

Never used, never diagnosed 242 38.84 99.84 (14.39) 100.64 (15.25) 0.05
Used, never diagnosed 479 49.48 102.32 (13.34) 101.25 (14.70) −0.07
1 diagnosis 80 70.00 96.40 (14.31) 94.78 (14.54) −0.11
2 diagnoses 35 62.86 102.14 (17.08) 99.67 (16.11) −0.17
3+ diagnoses 38 81.58 99.68 (13.53) 93.93 (13.32) −0.38
Persistence of regular cannabis use
Never used 242 38.84 99.84 (14.39) 100.64 (15.25) 0.05
Used, never regularly 508 50.59 102.27 (13.59) 101.24 (14.81) −0.07
Used regularly at 1 wave 47 72.34 101.42 (14.41) 98.45 (14.89) −0.20
Used regularly at 2 waves 36 63.89 95.28 (10.74) 93.26 (11.44) −0.13
Used regularly at 3+ waves 41 78.05 96.00 (16.06) 90.77 (13.88) −0.35

Means (SDs) are presented for child and adult full-scale IQ as a function of the number of study waves between ages 18 y and 38 y for
which study members met criteria for cannabis dependence or reported using cannabis on a regular basis (at least 4 d/wk). The last
column shows that study members with more persistent cannabis use showed greater IQ decline from childhood to adulthood.
*This coefficient indicates change in IQ from childhood to adulthood, with negative values indicating decreases in IQ. These change
scores are in SD units, with values of 0.20, 0.50, and 0.80 reflecting small, medium, and large changes, respectively.

E2658 | www.pnas.org/cgi/doi/10.1073/pnas.1206820109 Meier et al.

 PNAS PLUS
Table 2. IQ subtest changes
Never used, never Used, never 1 diagnosis, 2 diagnoses, 3+ diagnoses, Linear trend
IQ test/subtest diagnosed, n = 242 diagnosed, n = 479 n = 80 n = 35 n = 38 t test* P

Full-scale IQ 0.05 −0.07 −0.11 −0.17 −0.38 −4.45 <0.0001

−0.05 −0.13 −0.19 −0.31 −4.15 <0.0001

SEE COMMENTARY
Verbal IQ 0.02
Information subtest 0.05 −0.08 0.02 −0.25 −0.15 −2.40 0.0168
Similarities subtest 0.03 −0.05 −0.03 −0.19 −0.44 −2.78 0.0056
Vocabulary subtest 0.07 −0.05 −0.16 −0.16 −0.45 −3.67 0.0003
Arithmetic subtest −0.05 −0.07 −0.05 0.00 0.06 −0.73 0.47
Performance IQ 0.08 −0.08 −0.09 −0.08 −0.42 −2.84 0.0046
Digit symbol coding subtest 0.15 −0.09 −0.17 −0.23 −0.62 −5.60 <0.0001
Block design subtest −0.03 −0.07 −0.01 −0.11 0.02 −0.55 0.58
Picture completion subtest −0.01 −0.08 0.08 0.05 0.15 1.18 0.24

Mean change in IQ subtest scores from childhood to adulthood is presented in SD units as a function of the number of study waves between ages 18 y and
38 y for which a study member met criteria for cannabis dependence. These change scores can be interpreted as effect sizes, with values of 0.20, 0.50, and 0.80
reflecting small, medium, and large effects, respectively. Persistent cannabis dependence was associated with IQ decline for the majority of IQ subtests
administered in both childhood and adulthood, i.e., when each study member served as his or her own control.
*To test for a dose–response effect, we conducted an ordinary least-squares regression, estimating the linear trend controlling for sex.

scale IQ as a function of persistent cannabis dependence, ex- sequent analyses because full-scale IQ captures overall in-
cluding each of the aforementioned groups. We elected to show tellectual functioning. Fig. 1 shows that excluding each of these
results just for full-scale IQ for this analysis as well as all sub- groups of study members did not alter the initial finding; effect

Table 3. Five areas of mental function

Never used, Used, never
never diagnosed, diagnosed, 1 diagnosis, 2 diagnoses, 3+ diagnoses, Linear trend
Age 38 y neuropsychological tests n = 242 n = 479 n = 80 n = 35 n = 38 t test* P

PSYCHOLOGICAL AND
COGNITIVE SCIENCES
Tests of executive functions
WAIS-IV Working Memory Index 0.01 0.03 −0.16 −0.03 −0.16 −2.16 0.0311
Wechsler Memory Scale Months of the 0.24 0.01 −0.38 −0.23 −0.63 −5.24 <0.0001
Year Backward
Trail-Making Test B Time† −0.04 −0.03 0.16 0.08 0.19 1.15 0.25
CANTAB Rapid Visual Information 0.05 0.01 −0.02 −0.04 −0.45 −2.58 0.0100
Processing A Prime (Vigilance)
CANTAB Rapid Visual Information −0.02 0.01 0.06 0.04 −0.14 −0.05 0.96
Processing Total False Alarms†
Tests of memory
Rey Auditory Verbal Learning Total Recall 0.11 0.06 −0.26 −0.22 −0.48 −2.65 0.0081
Rey Auditory Verbal Learning Delayed Recall 0.14 0.02 −0.22 −0.28 −0.31 −2.11 0.0348
Wechsler Memory Scale Verbal Paired 0.07 0.06 −0.21 −0.21 −0.12 −1.48 0.14
Associates Total Recall
Wechsler Memory Scale Verbal Paired 0.07 0.06 −0.19 −0.15 −0.14 −1.07 0.29
Associates Delayed Recall
CANTAB Visual Paired Associates Learning 0.09 0.01 −0.06 −0.36 −0.10 −2.22 0.0270
First Trial Memory Score
CANTAB Visual Paired Associates Learning −0.07 −0.03 0.17 0.33 −0.06 1.41 0.16
Total Errors†
Tests of processing speed
WAIS-IV Processing Speed Index 0.14 0.03 −0.21 −0.05 −0.61 −3.64 0.0003
CANTAB Rapid Visual Information −0.13 0.04 0.06 −0.20 0.25 1.92 0.06
Processing Mean Latency†
CANTAB Reaction Time 5-Choice Reaction 0.19 −0.11 −0.13 −0.01 0.18 −0.38 0.71
Time†
Tests of perceptual reasoning
WAIS-IV Perceptual Reasoning Index 0.08 −0.02 0.07 −0.18 −0.12 −2.33 0.0202
Tests of verbal comprehension
WAIS-IV Verbal Comprehension Index 0.10 −0.01 −0.03 0.02 −0.23 −3.04 0.0025

Neuropsychological test scores at age 38 y are shown as a function of the number of study waves between ages 18 y and 38 y for which study members met
criteria for cannabis dependence. Scores are standardized means adjusted for baseline (childhood) full-scale IQ assessed before the onset of cannabis use.
These means can be interpreted as effect sizes, with values of 0.20, 0.50, and 0.80 reflecting small, medium, and large effects, respectively. Persistent cannabis
dependence was associated with impairment in each of the five areas of mental function. CANTAB, Cambridge Neuropsychological Test Automated Battery;
WAIS-IV, Wechsler Adult Intelligence Scale-IV.
*To test for a dose–response effect, we conducted an ordinary least-squares regression, estimating the linear trend controlling for childhood full-scale IQ and sex.
†
Higher score indicates worse performance.

Meier et al. PNAS | Published online August 27, 2012 | E2659

 3+ Diagnoses

2 Diagnoses

1 Diagnosis

Used, Never Diagnosed

Never Used, Never Diagnosed

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2

Full Birth Cohort (n=874) Excluding Those with Persistent Hard-Drug Dependence (n=7)
Excluding Past-24-Hour Cannabis Users (n=38) Excluding Those with Persistent Alcohol Dependence (n=53)
Excluding Past-Week Cannabis Users (n=89) Excluding Those with Schizophrenia (n=28)
Excluding Those with Persistent Tobacco Dependence (n=126)

Fig. 1. Ruling out alternative explanations. Shown is change in full-scale IQ (in SD units) from childhood to adulthood as a function of the number of study
waves between ages 18 y and 38 y for which a study member met criteria for cannabis dependence. Change scores are presented for the full birth cohort and
the cohort excluding (i) past 24-h cannabis users, (ii) past-week cannabis users, (iii) those with persistent tobacco dependence, (iv) those with persistent hard-
drug dependence, (v) those with persistent alcohol dependence, and (vi) those with lifetime schizophrenia. Persistent tobacco, hard-drug, and alcohol de-
pendence were each defined as dependence at three or more study waves. IQ decline could not be explained by other factors. Error bars = SEs.

sizes, representing within-person IQ change as a function of tobacco, hard-drug, or alcohol dependence), and schizophrenia
persistent cannabis dependence, remained virtually the same and remained statistically significant (t = −2.20, P = 0.0282).
remained statistically significant (see Table S2 for IQ subtests).
Furthermore, a multivariate regression of the effect of persistent Is Impairment Apparent Even After Controlling for Years of Education?
cannabis dependence on full-scale IQ decline, controlling for The linear effect of persistent cannabis dependence on change in
past 24-h cannabis use, persistent substance dependence (the full-scale IQ was significant before controlling for years of edu-
number of study waves for which study members diagnosed with cation (t = −4.45, P < 0.0001; Table 2, top row) and remained

Table 4. IQ decline after holding education constant

Never used, Used, never Linear trend
Sample never diagnosed diagnosed 1 diagnosis 2 diagnoses 3+ diagnoses t test* P

Full sample 0.05 (n = 242) −0.07 (n = 479) −0.11 (n = 80) −0.17 (n = 35) −0.38 (n = 38) −4.45 <0.0001
High-school education or less −0.03 (n = 59) −0.14 (n = 130) −0.16 (n = 43) −0.25 (n = 20) −0.48 (n = 26) −3.36 0.0009

Mean change in full-scale IQ from childhood to adulthood is presented in SD units as a function of the number of study waves between ages 18 y and 38 y
for which a study member met criteria for cannabis dependence. These change scores can be interpreted as effect sizes, with values of 0.20, 0.50, and 0.80
reflecting small, medium, and large effects, respectively. Change scores are presented for the full sample and for the sample of study members with a high-
school education or less. Persistent cannabis dependence was associated with IQ decline in the full sample and the sample of study members with a high-
school education or less.
*To test for a dose–response effect, we conducted an ordinary least-squares regression, estimating the linear trend controlling for sex.

E2660 | www.pnas.org/cgi/doi/10.1073/pnas.1206820109 Meier et al.

 PNAS PLUS
significant after controlling for years of education (t = −3.41, P = persistent users. This effect was concentrated among adolescent-
0.0007). Moreover, although fewer persistent cannabis users onset cannabis users, a finding consistent with results of several
pursued education after high school (χ2 = 63.94, P < 0.0001), studies showing executive functioning or verbal IQ deficits
among the subset with a high-school education or less, persistent among adolescent-onset but not adult-onset chronic cannabis
cannabis users experienced greater IQ decline (Table 4). users (8, 10, 14, 15), as well as studies showing impairment of

SEE COMMENTARY
learning, memory, and executive functions in samples of ado-
Does Cannabis-Associated Neuropsychological Impairment Translate lescent cannabis users (11–13, 32).
into Functional Problems in Daily Life? Informant reports of study The present study advances knowledge in five ways. First, by
members’ neuropsychological functioning were also obtained at investigating the association between persistent cannabis use and
age 38 y. Study members nominated people “who knew them neuropsychological functioning prospectively, we ruled out pre-
well.” These informants were mailed questionnaires and asked to morbid neuropsychological deficit as an explanation of the link
complete a checklist, including whether the study members had between persistent cannabis use and neuropsychological impair-
problems with their attention and memory over the past year. ment occurring after persistent use. Second, we showed that the
Table 5 shows mean informant-reported cognitive problems, impairment was global and detectable across five domains of
adjusted for childhood IQ, as a function of persistent cannabis neuropsychological functioning. Third, we showed that cannabis-
dependence. Informants reported observing significantly more associated neuropsychological decline did not occur solely be-
attention and memory problems among those with more per- cause cannabis users completed fewer years of education. Fourth,
sistent cannabis dependence. we showed that impairment was apparent to third-party inform-
ants and that persistent cannabis use interfered with everyday
Are Adolescent Cannabis Users Particularly Vulnerable? Adolescent- cognitive functioning. Fifth, we showed that, among adolescent-
onset users, who diagnosed with cannabis dependence before age onset former persistent cannabis users, impairment was still evi-
18 y, tended to become more persistent users, but Fig. 2 shows dent after cessation of use for 1 y or more. Collectively, these
that, after equating adolescent- and adult-onset cannabis users findings are consistent with speculation that cannabis use in ad-
on total number of cannabis-dependence diagnoses, adolescent- olescence, when the brain is undergoing critical development,
onset users showed greater IQ decline than adult-onset cannabis may have neurotoxic effects.
users. In fact, adult-onset cannabis users did not appear to ex- The study’s results must be interpreted in the context of its
perience IQ decline as a function of persistent cannabis use. limitations. First, although we were able to rule out a set of
Because it might be difficult to develop cannabis dependence plausible alternative explanations for the association between
before age 18 y, we also defined adolescent-onset cannabis use in persistent cannabis use and neuropsychological functioning, such
terms of weekly use before age 18 y [the correspondence be- as premorbid neuropsychological deficit and hard-drug and al-

PSYCHOLOGICAL AND
tween cannabis dependence before age 18 y and weekly use cohol dependence among persistent cannabis users, our data

COGNITIVE SCIENCES
before age 18 y was not perfect (κ = 0.64)]. Results of this cannot definitively attest to whether this association is causal.
analysis (Fig. S1) were similar. For example, there may be some unknown “third” variable that
could account for the findings. The data also cannot reveal the
What Is the Effect of Cessation of Cannabis Use? Given that ado- mechanism underlying the association between persistent can-
lescent-onset cannabis users exhibited marked IQ decline and nabis dependence and neuropsychological decline. One hy-
given speculation that this could represent a toxic effect of pothesis is that cannabis use in adolescence causes brain changes
cannabis on the developing brain, we examined the cessation that result in neuropsychological impairment. Several lines of
effect separately within adolescent-onset and adult-onset can- evidence support this possibility (24–31, 33, 34). First, puberty is
nabis users. Fig. 3 shows that, among adolescent-onset persistent a period of critical brain development, characterized by neuronal
cannabis users, within-person IQ decline was apparent regardless maturation and rearrangement processes (e.g., myelination,
of whether cannabis was used infrequently (median use = 14 d) synaptic pruning, dendritic plasticity) and the maturation of
or frequently (median use = 365 d) in the year before testing. In neurotransmitter systems (e.g., the endogenous cannabinoid
contrast, within-person IQ decline was not apparent among system), making the pubertal brain vulnerable to toxic insult
adult-onset persistent cannabis users who used cannabis in- (33). Second, cannabis administration in animals is associated
frequently (median use = 6 d) or frequently (median use = 365 d) with structural and functional brain differences, particularly in
in the year before testing. Thus, cessation of cannabis use did not hippocampal regions, with structural differences dependent on
fully restore neuropsychological functioning among adolescent- age and duration of exposure to cannabinoids (33). Third,
onset former persistent cannabis users. studies of human adolescents have shown structural and func-
tional brain differences associated with cannabis use (26, 29, 35).
Discussion Alternatively, persistent cannabis users may experience greater
Persistent cannabis use over 20 y was associated with neuro- neuropsychological decline relative to nonusers because they
psychological decline, and greater decline was evident for more receive less education. Our results suggest that cannabis-associ-

Table 5. Cognitive problems outside the laboratory

Never used, Used, never
never diagnosed, diagnosed, 1 diagnosis, 2 diagnoses, 3+ diagnoses, Linear trend
Age 38 y informant reports n = 228 n = 457 n = 71 n = 31 n = 35 t test* P
†
Informant-reported attention problems −0.21 −0.07 0.31 0.64 0.96 7.74 <0.0001
Informant-reported memory problems† −0.27 −0.03 0.38 0.78 0.75 7.65 <0.0001

Shown are informant reports of cognitive problems at age 38 y as a function of the number of study waves between ages 18 y and 38 y for which study
members met criteria for cannabis dependence. Scores are standardized means adjusted for baseline (childhood) full-scale IQ assessed before the onset of
cannabis use. These means can be interpreted as effect sizes, with values of 0.20, 0.50, and 0.80 reflecting small, medium, and large effects, respectively.
Cognitive problems among persistent cannabis users were apparent to the “naked-eye.”
*To test for a dose–response effect, we conducted an ordinary least-squares regression, estimating the linear trend controlling for childhood full-scale IQ and sex.
†
Higher score indicates worse everday problems.

Meier et al. PNAS | Published online August 27, 2012 | E2661

 0.4
1 Diagnosis 2 Diagnoses 3+ Diagnoses

0.2

(in standard deviation units)

Change in Full-Scale IQ
0

-0.2

-0.4

-0.6

-0.8 p = .44 p = .09 p = .02

Cannabis Not Cannabis Cannabis Not Cannabis Cannabis Not Cannabis

Dependent Dependent Dependent Dependent Dependent Dependent
Before Age 18 Before Age 18 Before Age 18 Before Age 18 Before Age 18 Before Age 18
(n=17) (n=57) (n=12) (n=21) (n=23) (n=14)

Fig. 2. Adolescent vulnerability. Shown is change in full-scale IQ (in SD units) from childhood to adulthood among study members with 1, 2, or 3+ diagnoses
of cannabis dependence as a function of age of onset of cannabis dependence. Individuals with adolescent-onset cannabis dependence (black bars) expe-
rienced greater IQ decline than individuals with adult-onset cannabis dependence (gray bars). IQ decline of approximately −0.55 SD units among individuals
with adolescent-onset cannabis dependence in the 3+ group represents a decline of 8 IQ points. Error bars = SEs.

ated neuropsychological decline does not occur solely for this cent-onset former persistent cannabis users who quit or re-
reason, because the association between persistent cannabis duced their use for 1 y or more suggests that neuropsycholo-
use and neuropsychological decline was still apparent after gical functioning is not fully restored in this time. Fifth, these
controlling for years of education. Notably, the aforementioned findings are limited to a cohort of individuals born in Dunedin,
processes are not mutually exclusive and may, in fact, be in- New Zealand in the 1970s. Notably, the prevalence of cannabis
terrelated. For example, the toxic effects of cannabis on the brain dependence is somewhat higher among New Zealanders than
may result in impaired neuropsychological functioning, poor Americans (39), but the potency of cannabis obtained from
academic performance, and subsequent school dropout, which police seizures in New Zealand is similar to that of cannabis in
then results in further neuropsychological decline. In this case, the United States (40, 41).
our statistical control for education in the analysis of the asso- Increasing efforts should be directed toward delaying the on-
ciation between persistent cannabis use and neuropsychol- set of cannabis use by young people, particularly given the recent
ogical decline is likely an overcontrol (36). trend of younger ages of cannabis-use initiation in the United
A second limitation is that we obtained information on past- States and evidence that fewer adolescents believe that cannabis
year cannabis dependence and self-reported frequency of use is associated with serious health risk (42). In the present
cannabis use with no external validation of use (e.g., biological study, the most persistent adolescent-onset cannabis users evi-
assays). Validation of cannabis use through laboratory meas- denced an average 8-point IQ decline from childhood to adult-
ures could have helped detect cannabis users who did not re- hood. Quitting, however, may have beneficial effects, preventing
port use. Underreporting of cannabis use due to concerns additional impairment for adolescent-onset users. Prevention
about admitting to using an illegal substance is unlikely, and policy efforts should focus on delivering to the public the
however, because study members, interviewed repeatedly over message that cannabis use during adolescence can have harmful
38 y about a number of illegal activities, have learned to trust effects on neuropsychological functioning, delaying the onset of
the Dunedin Study’s confidentiality guarantee. Moreover, any cannabis use at least until adulthood, and encouraging cessation
such misclassification would have mitigated against differ- of cannabis use particularly for those who began using cannabis
ences. Third, additional research is needed to define the in adolescence.
parameters of use sufficient to produce neuropsychological
impairment, such as the quantity, frequency, and age-of-onset Methods
of use. Our findings suggest that regular cannabis use before Participants. Participants are members of the Dunedin Multidisciplinary
age 18 y predicts impairment, but others have found effects Health and Development Study, a longitudinal investigation of the health
only for younger ages (10, 15). Given that the brain undergoes and behavior of a complete birth cohort of consecutive births between April
1, 1972, and March 31, 1973, in Dunedin, New Zealand. The cohort of 1,037
dynamic changes from the onset of puberty through early
children (91% of eligible births; 52% boys) was constituted at age 3 y. Cohort
adulthood (37, 38), this developmental period should be the families represent the full range of socioeconomic status in the general
focus of future research on the age(s) at which harm occurs. population of New Zealand’s South Island and are primarily of white Euro-
Fourth, additional research is needed to determine whether pean ancestry. Follow-up assessments were conducted with informed con-
cannabis-related neuropsychological impairment is reversible. sent at 5, 7, 9, 11, 13, 15, 18, 21, 26, 32, and most recently at 38 y of age,
Our finding of neuropsychological difficulties among adoles- when 96% of the 1,004 living study members underwent assessment in

E2662 | www.pnas.org/cgi/doi/10.1073/pnas.1206820109 Meier et al.

 PNAS PLUS
p = .03 p = .0002 p = .73 p = .11

110 110

105 105

SEE COMMENTARY
Full-Scale IQ

100

Full-Scale IQ
100
Child IQ Child IQ
95 95
Adult IQ Adult IQ
90 90

85 85

80 80
Infrequent Frequent Infrequent Frequent
Cannabis Cannabis Cannabis Cannabis
Use at Age Use at Age Use at Age Use at Age
38 38 38 38
(n=17) (n=19) (n=13) (n=20)

Adolescent-Onset (Used Cannabis Weekly Before Age 18) Adult-Onset (Did Not Use Cannabis Weekly Before Age 18)
Fig. 3. Postcessation IQ among former persistent cannabis users. This figure is restricted to persistent cannabis users, defined as study members with two or
more diagnoses of cannabis dependence. Shown is full-scale IQ in childhood and adulthood. IQ is plotted as a function of (i) age of onset of at least weekly
cannabis use and (ii) the frequency of cannabis use at age 38 y. Infrequent use was defined as weekly or less frequent use in the year preceding testing at age
38 y. Median use among infrequent and frequent adolescent-onset cannabis users was 14 (range: 0–52) and 365 (range: 100–365) d, respectively. Median use
among infrequent and frequent adult-onset cannabis users was 6 (range: 0–52) and 365 (range: 100–365) d, respectively. IQ decline was apparent even after
cessation of cannabis use for adolescent-onset former persistent cannabis users. Error bars = SEs.

PSYCHOLOGICAL AND
COGNITIVE SCIENCES
2010–2012. The Otago Ethics Committee approved each wave of the study. cannabis dependence and regular cannabis-use groups was high but not
Study members gave informed consent before participating. perfect (weighted κ = 0.77).
Because individuals with missing data at one wave tend to return to the The Dunedin Study uses past-year reporting to maximize validity and
study at some later wave(s), the attrition in the Dunedin Study has not been reliability of recall. A potential consequence is that individuals could have
cumulative, and reasons for missing assessments seem to be idiosyncratic rather experienced dependence only during a gap between the Study’s five 12-mo
than systematic. There was no evidence of differential attrition for cannabis- assessment windows and gone uncounted. Our “net” of 1-y assessments at
dependent individuals. For example, the 4% of study members who did not ages 18, 21, 26, 32, and 38 y captured all but four of the cohort members
participate at age 38 y were no more likely to have been cannabis dependent who reported receiving treatment for a drug-use problem between as-
at age 18 y than study members who did participate (F = 2.22, P = 0.14). sessment windows. Three of the four were hard-drug and alcohol de-
pendent, and the remaining person sought counseling for cannabis use
Measures. Cannabis use. Past-year cannabis dependence was assessed with only as part of a child custody dispute. As these four cohort members
the Diagnostic Interview Schedule (43, 44) at ages 18, 21, 26, 32, and 38 y reported cannabis use but not dependence, they were classified as “used
following criteria for the Diagnostic and Statistical Manual of Mental Dis- but never diagnosed.”
orders (DSM) (45, 46). Cohort members having missing data from three or Neuropsychological functioning. Intelligence was assessed in childhood at ages
more of the five study waves (ages 18, 21, 26, 32, and 38 y) were excluded 7, 9, 11, and 13 y, before the onset of cannabis use (only seven study members
when we defined our cannabis-exposure variables: 97% of living cohort reported trying cannabis by age 13 y), and again in adulthood at age 38 y. We
members were studied, composed of 83% of living study members with no report comparison of the Wechsler Intelligence Scale for Children-Revised
missing data points, 11% with one missing data point, and 3% with two (WISC-R) (47) and the WAIS-IV (48), both with M = 100 and SD = 15. At age
missing data points. Our main exposure, persistence of cannabis de- 38 y, additional neuropsychological tests were administered, including the
pendence, was defined as the total number of study waves out of five at Wechsler Memory Scale-III (WMS-III) (49), the Trail-Making Test (50), the
which a study member met criteria for cannabis dependence. Study mem- Cambridge Neuropsychological Test Automated Battery (CANTAB) (51), and
bers were grouped according to their number of dependence diagnoses: (i) the Rey Auditory Verbal Learning Test (52). Because the sample is a repre-
those who never used cannabis at any study wave and thus could not have sentative birth cohort, it formed its own norms. Table S3 provides further
become dependent, (ii) those who used cannabis at least once at one or details about each test. Each study member attended the research unit for
more study waves but never diagnosed, (iii) those who diagnosed at one an 8-h day of assessments. All testing occurred in the morning in two 50-min
wave, (iv) those who diagnosed at two waves, and (v) those who diagnosed counterbalanced sessions.
at three or more waves. Informant reports of study members’ neuropsychological functioning
Because there were some study members who used cannabis on a regular were also obtained at age 38 y. Study members nominated people who
basis but never met full criteria for a diagnosis of cannabis dependence, we knew them well. These informants were mailed questionnaires and asked to
repeated analyses using persistent regular cannabis use as the exposure. At complete a checklist, including whether the study members had problems
each of the five study waves between ages 18–38 y, study members self- with their attention and memory over the past year. The informant-reported
reported the total number of days (0–365) they used cannabis over the attention problems scale consisted of four items: “is easily distracted, gets
preceding year. Persistence of regular cannabis use was defined as the total sidetracked easily,” “can’t concentrate, mind wanders,” “tunes out instead
number of study waves out of five at which a study member reported using of focusing,” and “has difficulty organizing tasks that have many steps”
cannabis 4 d/wk or more (the majority of days in a week). Study members (internal consistency reliability = 0.79). The informant-reported memory
were grouped as those who (i) never used cannabis, (ii) used but never problems scale consisted of three items: “has problems with memory,”
regularly, (iii) used regularly at one wave, (iv) used regularly at two waves, “misplaces wallet, keys, eyeglasses, paperwork,” and “forgets to do errands,
and (v) used regularly at three or more waves. Correspondence between return calls, pay bills” (internal consistency reliability = 0.64).

Meier et al. PNAS | Published online August 27, 2012 | E2663

Control variables. Past 24-h cannabis use and past-week cannabis use were predicted neuropsychological test performance in adulthood (i.e., resi-
assessed at age 38 y on the day of neuropsychological testing. Persistent DSM dualized change scores).
(45, 46) tobacco, hard-drug, and alcohol dependence were assessed over the Tables 2–5 present the t tests associated with the regression coefficient
same 20-y period during which cannabis dependence was assessed, and the testing the linear effect of persistent cannabis use on change in neuro-
number of study waves during which study members diagnosed was coun- psychological functioning, under the hypothesis that more persistent
ted and used as covariates. For Fig. 1, persistent dependence was defined as cannabis use predicts greater decline in neuropsychological functioning.
having been diagnosed at three or more study waves. Research diagnoses of Change scores are presented in SD units as a function of persistence of
lifetime schizophrenia (53) are also reported.
cannabis use. These scores can be interpreted as effect sizes, with values of
0.20, 0.50, and 0.80 reflecting small, medium, and large change, respectively
Statistical Analysis. First, for the IQ test and subtests (47, 48) administered in
(54). Sex was included as a covariate in all statistical tests.
both childhood and adulthood, change scores were created by subtracting
the precannabis childhood IQ averaged across ages 7, 9, 11 and 13 y (or, for
ACKNOWLEDGMENTS. We thank the Dunedin Study members, their fami-
the seven members who reported trying cannabis by age 13 y, ages 7, 9, and
lies, the Dunedin Multidisciplinary Health and Development Research Unit
11 y) from postcannabis adulthood IQ. Negative scores indicate IQ decline. staff, and study founder Phil Silva. The Dunedin Multidisciplinary Health and
Ordinary least-squares linear regression was used to test whether persistent Development Research Unit is supported by the New Zealand Health
cannabis use (entered as a five-level independent variable, with each study Research Council. This research received support from UK Medical Research
member receiving a score ranging from 1 to 5) predicted amount of IQ Council Grants G0100527 and MR/K00381X/1, US National Institute on Aging
change. Second, for the neuropsychological tests administered only in Grant AG032282, US National Institute of Mental Health Grant MH077874,
adulthood, ordinary least-squares linear regression, including full-scale and US National Institute on Drug Abuse Grant P30 DA023026. Additional
childhood IQ as a covariate, was used to test whether persistent cannabis use support was provided by the Jacobs Foundation.

1. Murray RM, Morrison PD, Henquet C, Di Forti M (2007) Cannabis, the mind and 28. Rubino T, et al. (2009) Changes in hippocampal morphology and neuroplasticity
society: The hash realities. Nat Rev Neurosci 8:885–895. induced by adolescent THC treatment are associated with cognitive impairment in
2. Solowij N, Battisti R (2008) The chronic effects of cannabis on memory in humans: A adulthood. Hippocampus 19:763–772.
review. Curr Drug Abuse Rev 1:81–98. 29. Ashtari M, et al. (2011) Medial temporal structures and memory functions in
3. Solowij N, et al.; Marijuana Treatment Project Research Group (2002) Cognitive adolescents with heavy cannabis use. J Psychiatr Res 45:1055–1066.
functioning of long-term heavy cannabis users seeking treatment. JAMA 287:1123–1131. 30. Chan GCK, Hinds TR, Impey S, Storm DR (1998) Hippocampal neurotoxicity of Delta9-
4. Pope HG, Jr., Yurgelun-Todd D (1996) The residual cognitive effects of heavy tetrahydrocannabinol. J Neurosci 18:5322–5332.
marijuana use in college students. JAMA 275:521–527. 31. Schneider M, Schömig E, Leweke FM (2008) Acute and chronic cannabinoid treatment
5. Fletcher JM, et al. (1996) Cognitive correlates of long-term cannabis use in Costa Rican differentially affects recognition memory and social behavior in pubertal and adult
men. Arch Gen Psychiatry 53:1051–1057. rats. Addict Biol 13:345–357.
6. Bolla KI, Brown K, Eldreth D, Tate K, Cadet JL (2002) Dose-related neurocognitive 32. Hanson KL, et al. (2010) Longitudinal study of cognition among adolescent marijuana
effects of marijuana use. Neurology 59:1337–1343. users over three weeks of abstinence. Addict Behav 35:970–976.
7. Solowij N (1998) Cannabis and Cognitive Functioning (Cambridge Univ Press, 33. Jager G, Ramsey NF (2008) Long-term consequences of adolescent cannabis exposure
Cambridge, UK). on the development of cognition, brain structure and function: An overview of
8. Pope HG, Jr., et al. (2003) Early-onset cannabis use and cognitive deficits: What is the animal and human research. Curr Drug Abuse Rev 1:114–123.
nature of the association? Drug Alcohol Depend 69:303–310. 34. Scallet AC (1991) Neurotoxicology of cannabis and THC: A review of chronic exposure
9. Pope HG, Jr., Gruber AJ, Hudson JI, Huestis MA, Yurgelun-Todd D (2001) studies in animals. Pharmacol Biochem Behav 40:671–676.
Neuropsychological performance in long-term cannabis users. Arch Gen Psychiatry 58: 35. Jager G, Block RI, Luijten M, Ramsey NF (2010) Cannabis use and memory brain
909–915. function in adolescent boys: A cross-sectional multicenter functional magnetic
10. Gruber SA, Sagar KA, Dahlgren MK, Racine M, Lukas SE (2011) Age of onset of resonance imaging study. J Am Acad Child Adolesc Psychiatry 49:561–572, 572, e1–e3.
marijuana use and executive function. Psychol Addict Behav, 10.1037/a0026269. 36. Meehl PE (1971) High school yearbooks: A reply to Schwarz. J Abnorm Psychol 77:
11. Harvey MA, Sellman JD, Porter RJ, Frampton CM (2007) The relationship between 143–148.
non-acute adolescent cannabis use and cognition. Drug Alcohol Rev 26:309–319. 37. Giedd JN, et al. (1999) Brain development during childhood and adolescence: A
12. Medina KL, et al. (2007) Neuropsychological functioning in adolescent marijuana longitudinal MRI study. Nat Neurosci 2:861–863.
users: Subtle deficits detectable after a month of abstinence. J Int Neuropsychol Soc 38. Paus T (2005) Mapping brain maturation and cognitive development during
13:807–820. adolescence. Trends Cogn Sci 9:60–68.
13. Solowij N, et al. (2011) Verbal learning and memory in adolescent cannabis users, 39. Moffitt TE, et al. (2010) How common are common mental disorders? Evidence that
alcohol users and non-users. Psychopharmacology (Berl) 216:131–144. lifetime prevalence rates are doubled by prospective versus retrospective ascertainment.
14. Ehrenreich H, et al. (1999) Specific attentional dysfunction in adults following early Psychol Med 40:899–909.
start of cannabis use. Psychopharmacology (Berl) 142:295–301. 40. McLaren J, Swift W, Dillon P, Allsop S (2008) Cannabis potency and contamination:
15. Fontes MA, et al. (2011) Cannabis use before age 15 and subsequent executive A review of the literature. Addiction 103:1100–1109.
functioning. Br J Psychiatry 198:442–447. 41. Hall W, Swift W (2000) The THC content of cannabis in Australia: Evidence and
16. Messinis L, Kyprianidou A, Malefaki S, Papathanasopoulos P (2006) Neuropsychological implications. Aust N Z J Public Health 24:503–508.
deficits in long-term frequent cannabis users. Neurology 66:737–739. 42. Johnston LD, O’Malley PM, Bachman JG, Schulenberg JE (2011) Monitoring the Future
17. Giancola PR, Tarter RE (1999) Executive cognitive functioning and risk for substance National Survey Results on Drug Use, 1975-2010: Secondary School Students (Institute
abuse. Psychol Sci 10:203–205. for Social Research, University of Michigan, Ann Arbor, MI), Vol I.
18. Tarter RE, et al. (2003) Neurobehavioral disinhibition in childhood predicts early age 43. Robins LN, Helzer JE, Croughan J, Ratcliff KS (1981) National Institute of Mental
at onset of substance use disorder. Am J Psychiatry 160:1078–1085. Health Diagnostic Interview Schedule. Its history, characteristics, and validity. Arch
19. Ersche KD, et al. (2012) Abnormal brain structure implicated in stimulant drug Gen Psychiatry 38:381–389.
addiction. Science 335:601–604. 44. Robins LN, Cottler L, Bucholz KK, Compton W (1995) Diagnostic Interview Schedule
20. Verdejo-García A, Lawrence AJ, Clark L (2008) Impulsivity as a vulnerability marker for for DSM-IV (Washington Univ School of Medicine, St Louis, MO).
substance-use disorders: Review of findings from high-risk research, problem 45. American Psychiatric Association (1987) Diagnostic and Statistical Manual of Mental
gamblers and genetic association studies. Neurosci Biobehav Rev 32:777–810. Disorders (American Psychiatric Association, Washington, DC), 3rd Ed, Revised.
21. Fried PA, Watkinson B, Gray R (2005) Neurocognitive consequences of marihuana— 46. American Psychiatric Association (1994) Diagnostic and Statistical Manual of Mental
a comparison with pre-drug performance. Neurotoxicol Teratol 27:231–239. Disorders (American Psychiatric Association, Washington, DC), 4th Ed.
22. Brinch CN, Galloway TA (2011) Schooling in adolescence raises IQ scores. Proc Natl 47. Wechsler D (1974) Manual for the Wechsler Intelligence Scale for Children—Revised
Acad Sci USA 109:425–430. (Psychological Corporation, New York).
23. Horwood LJ, et al. (2010) Cannabis use and educational achievement: Findings from 48. Wechsler D (2008) Wechsler Adult Intelligence Scale (Pearson Assessment, San
three Australasian cohort studies. Drug Alcohol Depend 110:247–253. Antonio, TX), 4th Ed.
24. Schneider M, Koch M (2003) Chronic pubertal, but not adult chronic cannabinoid 49. Wechsler D (1997) Wechsler Memory Scale (Psychological Corporation, San Antonio,
treatment impairs sensorimotor gating, recognition memory, and the performance in TX), 3rd Ed.
a progressive ratio task in adult rats. Neuropsychopharmacology 28:1760–1769. 50. Army Individual Test Battery (1944) Manual and Directions for Scoring (War
25. Solowij N, Pesa N (2010) [Cognitive abnormalities and cannabis use]. Rev Bras Department, Adjutant General’s Office, Washington, DC).
Psiquiatr 32(Suppl 1):S31–S40. 51. Sahakian BJ, Owen AM (1992) Computerized assessment in neuropsychiatry using
26. Wilson W, et al. (2000) Brain morphological changes and early marijuana use: A CANTAB: Discussion paper. J R Soc Med 85:399–402.
magnetic resonance and positron emission tomography study. J Addict Dis 19:1–22. 52. Lezak MD (2004) Neuropsychological Assessment (Oxford Univ Press, New York), 4th Ed.
27. O’Shea M, Singh ME, McGregor IS, Mallet PE (2004) Chronic cannabinoid exposure 53. Reichenberg A, et al. (2010) Static and dynamic cognitive deficits in childhood
produces lasting memory impairment and increased anxiety in adolescent but not preceding adult schizophrenia: A 30-year study. Am J Psychiatry 167:160–169.
adult rats. J Psychopharmacol 18:502–508. 54. Cohen J (1992) A power primer. Psychol Bull 112:155–159.

E2664 | www.pnas.org/cgi/doi/10.1073/pnas.1206820109 Meier et al.