Introduction
Ultra-high-risk (UHR) individuals refer to a group of individuals predominantly characterized by attenuated psychotic symptoms and marked reductions in level of functioning (Yung et al. Reference Yung, Phillips, Yuen, Francey, McFarlane, Hallgren and McGorry2003, Reference Yung, Yuen, McGorry, Phillips, Kelly, Dell'Olio, Francey, Cosgrave, Killackey, Stanford, Godfrey and Buckby2005), but negative symptoms are also frequently reported (Meyer et al. Reference Meyer, Carrion, Cornblatt, Addington, Cadenhead, Cannon, McGlashan, Perkins, Tsuang, Walker, Woods, Heinssen and Seidman2014; Schlosser et al. Reference Schlosser, Campellone, Biagianti, Delucchi, Gard, Fulford, Stuart, Fisher, Loewy and Vinogradov2015).
Cerebral white matter (WM) microstructure can be examined in vivo by diffusion tensor imaging (DTI), which is a magnetic resonance imaging (MRI) technique. From DTI data various tensor invariants can be obtained: fractional anisotropy (FA) and mean diffusivity (MD) reflecting the degree of directionality and average magnitude of molecular displacement by diffusion, and axial diffusivity (AD) and radial diffusivity (RD) reflecting diffusion parallel and perpendicular to the preferred direction of diffusion (Basser & Pierpaoli, Reference Basser and Pierpaoli2011). Mode of anisotropy (MO) reflects the degree to which anisotropy is planar (due to e.g. crossing fibers) or linear (as in a single fiber population orientation) (Ennis & Kindlmann, Reference Ennis and Kindlmann2006; Kindlmann et al. Reference Kindlmann, Ennis, Whitaker and Westin2007; Wigand et al. Reference Wigand, Kubicki, Clemm von, Leicht, Karch, Eckbo, Pelavin, Hawley, Rujescu, Bouix, Shenton and Mulert2015).
DTI has been increasingly applied in studies of patients with first episode psychosis (FEP) and schizophrenia (FES). These studies have indicated widespread aberrations in WM microstructure mainly characterized by FA reductions in various regions such as in the corpus callosum (Cheung et al. Reference Cheung, Cheung, McAlonan, Deng, Wong, Yip, Tai, Khong, Sham and Chua2008; Gasparotti et al. Reference Gasparotti, Valsecchi, Carletti, Galluzzo, Liserre, Cesana and Sacchetti2009; Perez-Iglesias et al. Reference Perez-Iglesias, Tordesillas-Gutierrez, Barker, McGuire, Roiz-Santianez, Mata, de Lucas, Quintana, Vazquez-Barquero and Crespo-Facorro2010), superior longitudinal fasciculus (Perez-Iglesias et al. Reference Perez-Iglesias, Tordesillas-Gutierrez, Barker, McGuire, Roiz-Santianez, Mata, de Lucas, Quintana, Vazquez-Barquero and Crespo-Facorro2010; Guo et al. Reference Guo, Liu, Liu, Gao, Xiao, Chen and Zhao2012), inferior longitudinal fasciculus (Cheung et al. Reference Cheung, Cheung, McAlonan, Deng, Wong, Yip, Tai, Khong, Sham and Chua2008; Perez-Iglesias et al. Reference Perez-Iglesias, Tordesillas-Gutierrez, Barker, McGuire, Roiz-Santianez, Mata, de Lucas, Quintana, Vazquez-Barquero and Crespo-Facorro2010; Ebdrup et al. Reference Ebdrup, Raghava, Nielsen, Rostrup and Glenthoj2016), internal capsule including the corticospinal tracts and the anterior and superior thalamic radiations (Cheung et al. Reference Cheung, Cheung, McAlonan, Deng, Wong, Yip, Tai, Khong, Sham and Chua2008; Perez-Iglesias et al. Reference Perez-Iglesias, Tordesillas-Gutierrez, Barker, McGuire, Roiz-Santianez, Mata, de Lucas, Quintana, Vazquez-Barquero and Crespo-Facorro2010; Guo et al. Reference Guo, Liu, Liu, Gao, Xiao, Chen and Zhao2012; Filippi et al. Reference Filippi, Canu, Gasparotti, Agosta, Valsecchi, Lodoli, Galluzzo, Comi and Sacchetti2014; Ebdrup et al. Reference Ebdrup, Raghava, Nielsen, Rostrup and Glenthoj2016), external capsule (Guo et al. Reference Guo, Liu, Liu, Gao, Xiao, Chen and Zhao2012), and fronto-occipital fasciculus (Cheung et al. Reference Cheung, Cheung, McAlonan, Deng, Wong, Yip, Tai, Khong, Sham and Chua2008).
Fewer studies have investigated WM structure in UHR individuals (Peters et al. Reference Peters, de Haan, Dekker, Blaas, Becker, Dingemans, Akkerman, Majoie, van Amelsvoort, den Heeten and Linszen2008, Reference Peters, Schmitz, Dingemans, van Amelsvoort, Linszen, de Haan, Majoie and den Heeten2009, Reference Peters, Dingemans, Dekker, Blaas, Akkerman, van Amelsvoort, Majoie, den Heeten, Linszen and de Haan2010; Karlsgodt et al. Reference Karlsgodt, Niendam, Bearden and Cannon2009; Bloemen et al. Reference Bloemen, de Koning, Schmitz, Nieman, Becker, de Haan, Dingemans, Linszen and van Amelsvoort2010; Carletti et al. Reference Carletti, Woolley, Bhattacharyya, Perez-Iglesias, Fusar, Valmaggia, Broome, Bramon, Johns, Giampietro, Williams, Barker and McGuire2012; Pettersson-Yeo et al. Reference Pettersson-Yeo, Benetti, Marquand, Dell'acqua, Williams, Allen, Prata, McGuire and Mechelli2013; Epstein et al. Reference Epstein, Cullen, Mueller, Robinson, Lee and Kumra2014; Mittal et al. Reference Mittal, Dean, Bernard, Orr, Pelletier-Baldelli, Carol, Gupta, Turner, Leopold, Robustelli and Millman2014; von Hohenberg et al. Reference von Hohenberg, Pasternak, Kubicki, Ballinger, Vu, Swisher, Green, Giwerc, Dahlben, Goldstein, Woo, Petryshen, Mesholam-Gately, Woodberry, Thermenos, Mulert, McCarley, Seidman and Shenton2014; Bernard et al. Reference Bernard, Orr and Mittal2015; Katagiri et al. Reference Katagiri, Pantelis, Nemoto, Zalesky, Hori, Shimoji, Saito, Ito, Dwyer, Fukunaga, Morita, Tsujino, Yamaguchi, Shiraga, Aoki and Mizuno2015; Schmidt et al. Reference Schmidt, Lenz, Smieskova, Harrisberger, Walter, Riecher-Rossler, Simon, Lang, McGuire, Fusar-Poli and Borgwardt2015; Cooper et al. Reference Cooper, Alm, Olson and Ellman2016; Rigucci et al. Reference Rigucci, Santi, Corigliano, Imola, Rossi-Espagnet, Mancinelli, De Pisa, Manfredi, Bozzao, Carducci, Girardi and Comparelli2016; Wang et al. Reference Wang, Ji, Hong, Poh, Krishnan, Lee, Rekhi, Keefe, Adcock, Wood, Fornito, Pasternak, Chee and Zhou2016). Similar to findings in schizophrenia patients most of the studies find that UHR individuals have widespread regional FA reductions (Karlsgodt et al. Reference Karlsgodt, Niendam, Bearden and Cannon2009; Peters et al. Reference Peters, Schmitz, Dingemans, van Amelsvoort, Linszen, de Haan, Majoie and den Heeten2009; Bloemen et al. Reference Bloemen, de Koning, Schmitz, Nieman, Becker, de Haan, Dingemans, Linszen and van Amelsvoort2010; Carletti et al. Reference Carletti, Woolley, Bhattacharyya, Perez-Iglesias, Fusar, Valmaggia, Broome, Bramon, Johns, Giampietro, Williams, Barker and McGuire2012; Epstein et al. Reference Epstein, Cullen, Mueller, Robinson, Lee and Kumra2014; Bernard et al. Reference Bernard, Orr and Mittal2015; Katagiri et al. Reference Katagiri, Pantelis, Nemoto, Zalesky, Hori, Shimoji, Saito, Ito, Dwyer, Fukunaga, Morita, Tsujino, Yamaguchi, Shiraga, Aoki and Mizuno2015; Cooper et al. Reference Cooper, Alm, Olson and Ellman2016; Rigucci et al. Reference Rigucci, Santi, Corigliano, Imola, Rossi-Espagnet, Mancinelli, De Pisa, Manfredi, Bozzao, Carducci, Girardi and Comparelli2016; Wang et al. Reference Wang, Ji, Hong, Poh, Krishnan, Lee, Rekhi, Keefe, Adcock, Wood, Fornito, Pasternak, Chee and Zhou2016), although a few studies find no aberrations in FA (Peters et al. Reference Peters, de Haan, Dekker, Blaas, Becker, Dingemans, Akkerman, Majoie, van Amelsvoort, den Heeten and Linszen2008, Reference Peters, Dingemans, Dekker, Blaas, Akkerman, van Amelsvoort, Majoie, den Heeten, Linszen and de Haan2010; Mittal et al. Reference Mittal, Dean, Bernard, Orr, Pelletier-Baldelli, Carol, Gupta, Turner, Leopold, Robustelli and Millman2014; von Hohenberg et al. Reference von Hohenberg, Pasternak, Kubicki, Ballinger, Vu, Swisher, Green, Giwerc, Dahlben, Goldstein, Woo, Petryshen, Mesholam-Gately, Woodberry, Thermenos, Mulert, McCarley, Seidman and Shenton2014) or even an increase in FA (Schmidt et al. Reference Schmidt, Lenz, Smieskova, Harrisberger, Walter, Riecher-Rossler, Simon, Lang, McGuire, Fusar-Poli and Borgwardt2015) in UHR individuals compared with healthy controls (HCs). Reduced FA in the superior longitudinal fasciculus appears to be the most consistent finding in UHR studies (Karlsgodt et al. Reference Karlsgodt, Niendam, Bearden and Cannon2009; Bloemen et al. Reference Bloemen, de Koning, Schmitz, Nieman, Becker, de Haan, Dingemans, Linszen and van Amelsvoort2010; Carletti et al. Reference Carletti, Woolley, Bhattacharyya, Perez-Iglesias, Fusar, Valmaggia, Broome, Bramon, Johns, Giampietro, Williams, Barker and McGuire2012; von Hohenberg et al. Reference von Hohenberg, Pasternak, Kubicki, Ballinger, Vu, Swisher, Green, Giwerc, Dahlben, Goldstein, Woo, Petryshen, Mesholam-Gately, Woodberry, Thermenos, Mulert, McCarley, Seidman and Shenton2014; Rigucci et al. Reference Rigucci, Santi, Corigliano, Imola, Rossi-Espagnet, Mancinelli, De Pisa, Manfredi, Bozzao, Carducci, Girardi and Comparelli2016; Wang et al. Reference Wang, Ji, Hong, Poh, Krishnan, Lee, Rekhi, Keefe, Adcock, Wood, Fornito, Pasternak, Chee and Zhou2016)
Currently it is unclear to what extent WM aberrations (if present) are associated with clinical symptoms and/or level of functioning in UHR individuals. Several studies have found a negative correlation between FA and severity of positive symptoms in UHR individuals and individuals experiencing sub-threshold psychotic symptoms (Bloemen et al. Reference Bloemen, de Koning, Schmitz, Nieman, Becker, de Haan, Dingemans, Linszen and van Amelsvoort2010; Cooper et al. Reference Cooper, Alm, Olson and Ellman2016; DeRosse et al. Reference DeRosse, Ikuta, Karlsgodt, Peters, Gopin, Szeszko and Malhotra2016; Wang et al. Reference Wang, Ji, Hong, Poh, Krishnan, Lee, Rekhi, Keefe, Adcock, Wood, Fornito, Pasternak, Chee and Zhou2016); however, one study found the inverse correlation in some of the same areas (Schmidt et al. Reference Schmidt, Lenz, Smieskova, Harrisberger, Walter, Riecher-Rossler, Simon, Lang, McGuire, Fusar-Poli and Borgwardt2015). Moreover, lower FA at baseline has been found to predict deterioration in level of functioning at follow-up (Karlsgodt et al. Reference Karlsgodt, Niendam, Bearden and Cannon2009) and a longitudinal increase in FA has been correlated to an improvement in severity of positive symptoms in UHR individuals (Katagiri et al. Reference Katagiri, Pantelis, Nemoto, Zalesky, Hori, Shimoji, Saito, Ito, Dwyer, Fukunaga, Morita, Tsujino, Yamaguchi, Shiraga, Aoki and Mizuno2015). In patients with schizophrenia findings are heterogeneous with findings of FA being both negatively and positively correlated to severity of positive (Hubl et al. Reference Hubl, Koenig, Strik, Federspiel, Kreis, Boesch, Maier, Schroth, Lovblad and Dierks2004; Seok et al. Reference Seok, Park, Chun, Lee, Cho, Kwon and Kim2007; Shergill et al. Reference Shergill, Kanaan, Chitnis, O'Daly, Jones, Frangou, Williams, Howard, Barker, Murray and McGuire2007; Skelly et al. Reference Skelly, Calhoun, Meda, Kim, Mathalon and Pearlson2008; Szeszko et al. Reference Szeszko, Robinson, Ashtari, Vogel, Betensky, Sevy, Ardekani, Lencz, Malhotra, McCormack, Miller, Lim, Gunduz-Bruce, Kane and Bilder2008; Whitford et al. Reference Whitford, Kubicki, Schneiderman, O'Donnell, King, Alvarado, Khan, Markant, Nestor, Niznikiewicz, McCarley, Westin and Shenton2010; Catani et al. Reference Catani, Craig, Forkel, Kanaan, Picchioni, Toulopoulou, Shergill, Williams, Murphy and McGuire2011; Cheung et al. Reference Cheung, Chiu, Law, Cheung, Hui, Chan, Sham, Deng, Tai, Khong, McAlonan, Chua and Chen2011; Choi et al. Reference Choi, Kubicki, Whitford, Alvarado, Terry, Niznikiewicz, McCarley, Kwon and Shenton2011; Cui et al. Reference Cui, Chen, Deng, Huang, Li, Ma, Huang, Jiang, Wang, Wang, Collier, Gong and Li2011; Mulert & Scarr, Reference Mulert and Scarr2012; Boos et al. Reference Boos, Mandl, van Haren, Cahn, van Baal, Kahn and Hulshoff Pol2013; Lee et al. Reference Lee, Kubicki, Asami, Seidman, Goldstein, Mesholam-Gately, McCarley and Shenton2013; Hatton et al. Reference Hatton, Lagopoulos, Hermens, Hickie, Scott and Bennett2014; Caprihan et al. Reference Caprihan, Jones, Chen, Lemke, Abbott, Qualls, Canive, Gasparovic and Bustillo2015; Ohtani et al. Reference Ohtani, Bouix, Lyall, Hosokawa, Saito, Melonakos, Westin, Seidman, Goldstein, Mesholam-Gately, Petryshen, Wojcik and Kubicki2015; Ebdrup et al. Reference Ebdrup, Raghava, Nielsen, Rostrup and Glenthoj2016; Seitz et al. Reference Seitz, Zuo, Lyall, Makris, Kikinis, Bouix, Pasternak, Fredman, Duskin, Goldstein, Petryshen, Mesholam-Gately, Wojcik, McCarley, Seidman, Shenton, Koerte and Kubicki2016; Zhang et al. Reference Zhang, Fan, Chen, Tan, Tan, Hu, Salas, Kosten, Zunta-Soares and Soares2016) and negative (Wolkin et al. Reference Wolkin, Choi, Szilagyi, Sanfilipo, Rotrosen and Lim2003; Shin et al. Reference Shin, Kwon, Ha, Park, Kim, Hong, Moon, Lee, Kim, Kim and Chung2006; Skelly et al. Reference Skelly, Calhoun, Meda, Kim, Mathalon and Pearlson2008; Szeszko et al. Reference Szeszko, Robinson, Ashtari, Vogel, Betensky, Sevy, Ardekani, Lencz, Malhotra, McCormack, Miller, Lim, Gunduz-Bruce, Kane and Bilder2008; Lee et al. Reference Lee, Kubicki, Asami, Seidman, Goldstein, Mesholam-Gately, McCarley and Shenton2013; Asami et al. Reference Asami, Hyuk, Bouix, Rathi, Whitford, Niznikiewicz, Nestor, McCarley, Shenton and Kubicki2014; Ohtani et al. Reference Ohtani, Bouix, Lyall, Hosokawa, Saito, Melonakos, Westin, Seidman, Goldstein, Mesholam-Gately, Petryshen, Wojcik and Kubicki2015; Sun et al. Reference Sun, Lui, Yao, Deng, Xiao, Zhang, Huang, Hu, Bi, Li, Sweeney and Gong2015; Zeng et al. Reference Zeng, Ardekani, Tang, Zhang, Zhao, Cui, Fan, Zhuo, Li, Xu, Goff and Wang2016) symptoms.
Previous UHR studies in WM microstructure have employed univariate analysis investigating each diffusion parameter separately. However, it is likely that the complex pathological phenomena altering the tissue microstructure will affect several diffusion parameters simultaneously. In the present study, we are extending previous findings by doing joint analysis of diffusion parameters using multivariate analysis, which might provide additional information than the widely used univariate analysis (Konukoglu et al. Reference Konukoglu, Coutu, Salat and Fischl2016). To our knowledge this is the first study to explore multivariate patterns of WM microstructural pathology in UHR individuals and associating the patterns to clinical symptoms and level of functioning.
The clinical challenge of recruitment of UHR individuals is reflected in generally small samples sizes, which further challenge the application of conventional statistical procedures. Moreover, DTI data provide multiple interrelated measures, which require careful attention to problems with multiple comparisons, especially when interpreting voxelwise whole brain analyses. Multivariate statistical analyses, such as the partial least-square correlation (PLSC) procedure, may be superior to conventional univariate statistics to handle small sample sizes and interrelated measures (McIntosh et al. Reference McIntosh, Bookstein, Haxby and Grady1996; Nestor et al. Reference Nestor, O'Donnell, McCarley, Niznikiewicz, Barnard, Jen, Bookstein and Shenton2002; McIntosh & Lobaugh, Reference McIntosh and Lobaugh2004; Krishnan et al. Reference Krishnan, Williams, McIntosh and Abdi2011).
Hypothesis
We hypothesized different patterns of WM microstructure in UHR individuals compared with HCs, with UHR individuals showing a pattern of low FA, high RD, and unchanged or low AD. We also hypothesized that clinical symptoms and the level of functioning would be associated with patterns of WM microstructure within the UHR individuals.
We explored this using a multivariate approach since the complex phenomena of UHR is more likely associated to patterns of diffusion aberrations than to strongly localized effects (Konukoglu et al. Reference Konukoglu, Coutu, Salat and Fischl2016). We used bootstrap tests to visualize the most reliable spatial regions showing these WM patterns. However, interpretation of the spatial localization should be done with caution since the PLSC uses the entire set of voxels as a single multivariate representation. We therefore did additional univariate analysis exploring voxelwise group differences in each DTI parameter, and correlating each clinical measure to the DTI parameters separately. We also tried to quantify the PLSC findings using the spatial regions found by PLSC as regions of interest (ROIs) and compared mean values of DTI parameters in these regions.
Method
The study was conducted in accordance with the declaration of Helsinki and approved by the Regional Ethics Committee of the Capital Region, Denmark (H-D-2009-013). Participants signed informed consent.
Participants
As part of an UHR project at Mental Health Centre Copenhagen, UHR individuals aged 18–40 years were recruited between December 2008 and December 2014 from psychiatric in- and outpatient services in the Capital Region of Denmark. Individuals had to be help-seeking and fulfill criteria for one or more of the three UHR categories as assessed by the Comprehensive Assessment of At-Risk Mental States (CAARMS) (Yung et al. Reference Yung, Phillips, Yuen, Francey, McFarlane, Hallgren and McGorry2003, Reference Yung, Yuen, McGorry, Phillips, Kelly, Dell'Olio, Francey, Cosgrave, Killackey, Stanford, Godfrey and Buckby2005): (1) attenuated psychotic symptoms (APS); (2) brief limited intermittent psychotic symptoms (BLIPS); (3) trait and state (schizotypal personality disorder or a history of psychosis in a first-degree relative) (TS). An additional inclusion criterion was a decline in functioning (⩾30% drop for at least one month), or sustained low functioning (⩽50 for at least a year) assessed with the Social and Occupational Functioning Assessment Scale (SOFAS) (Goldman et al. Reference Goldman, Skodol and Lave1992; Hilsenroth et al. Reference Hilsenroth, Ackerman, Blagys, Baumann, Baity, Smith, Price, Smith, Heindselman, Mount and Holdwick2000).
Exclusion criteria were: organic brain disease, a diagnosis of a serious developmental disorder (e.g. Asperger's syndrome), IQ < 70 (as measured by the Wechsler Adult Intelligence Scale – Third Edition (WAIS-III)) (Wechsler, Reference Wechsler1997), psychiatric symptoms explained by acute intoxication or a physical illness with psychotropic effect, past history of a psychotic episode of more than one week, current treatment with mood stabilisers or methylphenidate, antipsychotic medication exceeding a total lifetime dose of 50 mg haloperidol or equivalent antipsychotic medication (Andreasen et al. Reference Andreasen, Pressler, Nopoulos, Miller and Ho2010), and pregnancy (tested with a urine HCG pregnancy test).
HCs were included between April 2009 and May 2014 as part of the Pan European Collaboration Antipsychotic-Naive Studies at Psychiatric Centre Glostrup. HCs had no first degree relatives with a psychotic disorder and underwent psychopathology assessments using the Schedules for Clinical Assessment in Neuropsychiatry Version 2.1 (Wing et al. Reference Wing, Babor, Brugha, Burke, Cooper, Giel, Jablenski, Regier and Sartorius1990) to ensure that they had no current or previous psychiatric diagnoses including substance abuse or dependency. The HCs were matched to the UHR individuals on gender and age (±3 years).
None of the participants had a history of major head injury resulting in loss of consciousness, and MRI scans were without overt pathology as evaluated by a trained neuro-radiologist.
Clinical measures
All participants underwent comprehensive assessments at baseline including socio-demographics, psychopathology, level of functioning, and MRI.
The Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) (First et al. Reference First, Spitzer, Gibbon and Williams2001) and part of the Structured Clinical Interview for DSM-IV Axis II Disorders (SCID-II) (First et al. Reference First, Spitzer, Gibbon, Williams and Benjamin1994) were used to diagnostically assess UHR individuals. All assessors (authors KK, DN, and LR) were certified SCID interviewers.
Positive symptoms were assessed with CAARMS (Yung et al. Reference Yung, Phillips, Yuen, Francey, McFarlane, Hallgren and McGorry2003, Reference Yung, Yuen, McGorry, Phillips, Kelly, Dell'Olio, Francey, Cosgrave, Killackey, Stanford, Godfrey and Buckby2005), and all assessors had undergone formal CAARMS training. Negative symptoms were assessed with Scale for the Assessment of Negative Symptoms (SANS) (Andreasen, Reference Andreasen1982). Level of functioning was assessed with the Social and Occupational Functioning Assessment Scale (SOFAS) (Goldman et al. Reference Goldman, Skodol and Lave1992; Hilsenroth et al. Reference Hilsenroth, Ackerman, Blagys, Baumann, Baity, Smith, Price, Smith, Heindselman, Mount and Holdwick2000). Handedness was assessed using the Edinburgh Handedness Inventory (Oldfield, Reference Oldfield1971).
Data on premorbid adjustment and systemic oxidative DNA and RNA damage in part of the UHR cohort has previously been published (Dannevang et al. Reference Dannevang, Randers, Gondan, Krakauer, Nordholm and Nordentoft2016; Nordholm et al. Reference Nordholm, Poulsen, Hjorthoj, Randers, Nielsen, Wulff, Krakauer, Norbak-Emig, Henriksen, Glenthoj and Nordentoft2016).
Image acquisition and processing
All MRIs were acquired in a 3 Tesla scanner (Philips Healthcare, Best, The Netherlands), using an 8-channel SENSE head coil (Invivo, Orlando, Florida, USA). Diffusion-weighted (DW) images were acquired using single shot spin-echo echoplanar imaging and a total of 31 different diffusion encodings: one diffusion un-weighted (b = 0 s/mm2) and 30 diffusion weighted (b = 1000 s/mm2) non-collinear directions. Acquired matrix size = 128 × 128 × 75; voxel dimensions = 1.88 × 1.88 × 2 mm3 (slice gap = 0); TR = 7035 ms; TE = 68 ms; flip angle = 90°. Acquisition of each volume was repeated with opposite phase encoding direction enabling correction for susceptibility distortions (Andersson & Skare, Reference Andersson and Skare2002; Andersson et al. Reference Andersson, Skare and Ashburner2003). Foam-pads inside the head coil minimized head movement during scanning.
Tools from the FSL FMRIB software library v5.0.9 were used for image processing (Smith et al. Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg, Bannister, De Luca, Drobnjak, Flitney, Niazy, Saunders, Vickers, Zhang, De Stefano, Brady and Matthews2004; Woolrich et al. Reference Woolrich, Jbabdi, Patenaude, Chappell, Makni, Behrens, Beckmann, Jenkinson and Smith2009; Jenkinson et al. Reference Jenkinson, Beckmann, Behrens, Woolrich and Smith2012). DW images were corrected for eddy current and head motion (Reese et al. Reference Reese, Heid, Weisskoff and Wedeen2003; Andersson & Sotiropoulos, Reference Andersson and Sotiropoulos2015), and absolute and relative head motion measures were estimated. Next, a susceptibility artifact correction was performed (Andersson & Skare, Reference Andersson and Skare2002). Motion corrected data were interpolated to 1.33 × 1.33 × 2 voxel dimensions using sinc interpolation and non-brain tissue was removed with brain extraction tool (BET) (Smith, Reference Smith2002). DTIFIT was used to calculate the diffusion parameter maps of FA, AD, RD, and MO and residual maps. The tract-based spatial statistics (TBSS) tool in FSL (Smith et al. Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg, Bannister, De Luca, Drobnjak, Flitney, Niazy, Saunders, Vickers, Zhang, De Stefano, Brady and Matthews2004, Reference Smith, Jenkinson, Johansen-Berg, Rueckert, Nichols, Mackay, Watkins, Ciccarelli, Cader, Matthews and Behrens2006) was used to align FA data using the nonlinear image registration tool (FNIRT) (Andersson et al. Reference Andersson, Jenkinson and Smith2007a , Reference Andersson, Jenkinson and Smith b ). The resulting mean FA image (threshold of 0.2) was thinned to create a mean FA skeleton, which represented the centers of all tracts common to the group. Next, the nearest maximum FA values of each registered FA image were projected onto the mean FA skeleton. AD, RD, and MO were also skeletonized by projecting onto the FA skeleton. The skeletonized images were corrected for absolute and relative head motion (in scanner), age, gender, parental socioeconomic status, antipsychotic medication, tobacco smoking, alcohol- and drug use using fsl_glm, and the residuals were z-transformed by subtracting the voxelwise mean and dividing by the voxelwise standard deviation across UHR individuals and HCs (Geladi & Kowalski, Reference Geladi and Kowalski1986).
MRI quality metrics
Quality control was done by visually inspecting all DW images slice by slice before processing and all squared residual maps that were created after calculating the DTI derived parameters maps. Additionally, four quality metrics [mean voxel intensity outlier count (MEANVOX), maximum voxel intensity outlier count (MAXVOX), mean relative head motion, and temporal signal-to-noise ratio (TSNR)] were computed from each DW image using a quality assessment method (Roalf et al. Reference Roalf, Quarmley, Elliott, Satterthwaite, Vandekar, Ruparel, Gennatas, Calkins, Moore, Hopson, Prabhakaran, Jackson, Verma, Hakonarson, Gur and Gur2016). Accordingly, these metrics were compared with standardized values in three different quality groups. In general, the quality metrics in the present study ranged between the ‘good’ and ‘excellent’ quality assessment group. For details see online Supplementary Table S1. Differences in absolute- and relative head motion (in scanner) summary measures between UHR individuals and HCs are provided in Table 1.
Demographics and clinical characteristics of participants
Table 1 shows demographics and clinical characteristics of participants
Drug use, Cannabis, opioids, hallucinogens, stimulants, other illicit drugs; Medication, UHR Individuals could be taking a combination of the listed compounds; APS, attenuated psychotic symptoms; BLIPS, brief limited intermittent psychotic symptoms; TS, trait and state; CAARMS, Comprehensive Assessment of At-Risk Mental States; SOFAS, Social and Occupational Functioning Assessment Scale; SANS, Scale for the Assessment of Negative Symptoms
a Fisher exact test
b Mann Whitney U
c Pearsons χ2
d Independent samples t test.
Statistical analysis on sociodemographic and clinical data
Statistical Package for the Social Sciences software (version 22.0, SPSS Inc., USA) was used to analyze demographic and clinical data. Distribution of continuous data was tested for normality with the Kolmogorov-Smirnov test. Since data on age, years of education, height, and weight were not normally distributed group comparisons were performed non-parametrically using the Mann–Whitney U test. Data on absolute- and relative head motion summary measures were normally distributed and analyzed with independent samples t test. Ordinal data such as tobacco smoking, alcohol and drug use were tested using the Mann–Whitney U test. Handedness was assessed using Pearson's χ2 test. Gender differences and parental socioeconomic status were assessed using Fisher's exact test.
A positive symptom severity score was calculated as the summed scores of the product of global rating scale scores (0–6) and frequency (0–6) scores of the four subscales of CAARMS (Morrison et al. Reference Morrison, French, Stewart, Birchwood, Fowler, Gumley, Jones, Bentall, Lewis, Murray, Patterson, Brunet, Conroy, Parker, Reilly, Byrne, Davies and Dunn2012). A negative symptom severity score was calculated by averaging the global scores on the first four items of the SANS leaving out the attention global score (Arndt et al. Reference Arndt, Andreasen, Flaum, Miller and Nopoulos1995).
Statistical analysis of MRI data using multivariate PLSC
Multivariate Partial Least Squares correlation analysis (PLSC) [graphical user interface, Rotman Research Institute, Baycrest Centre, University of Toronto; downloaded from http://www.rotman-baycrest.on.ca/pls/ running on MATLAB R2013b (MathWorks, Inc., Natick, Massachusetts, USA)] was used to explore group differences in DTI parameters (i.e. FA, AD, RD, and MO at each voxel) between UHR individuals and HCs and to analyze co-variations between DTI parameters and clinical measures (i.e. CAARMS, SANS, and SOFAS) within the UHR individuals.
In the group analyses ‘mean-centered task PLSC’ (Krishnan et al. Reference Krishnan, Williams, McIntosh and Abdi2011; Guitart-Masip et al. Reference Guitart-Masip, Salami, Garrett, Rieckmann, Lindenberger and Backman2016) was employed. DTI parameters of both UHR individuals and HCs were stored in a matrix denoted X, with rows representing participants and DTI parameters and columns representing spatial locations (i.e. voxels) within a brain mask. Using task PLSC a matrix R was constructed, so that each row contained the average of one parameter within one group. In the present case, R therefore had eight rows (four DTI parameters × two groups). After singular value decomposition (SVD) latent variables (LVs) were calculated by projecting the left and right singular vectors onto the X and design matrices. This produced eight pairs of LVs with corresponding singular values [representing the proportion of the covariance accounted for by that given LV (the cross-block covariance)] and voxel saliences (indicating the degree to which each voxel was related to the effect characterized by that LV).
In the analysis exploring co-variations within UHR individuals the DTI parameters and clinical measures were considered two different data-blocks, each represented in matrix form. The DTI data block was represented by a matrix X, which was structured as the one used in the group analysis however only including the UHR individuals. The clinical measures were represented by a matrix Y, in which rows represented UHR individuals and columns represented clinical measures. The clinical measures were repeated to obtain the same no. of rows as in the DTI data block. Using ‘behavior PLSC’ a cross-correlation matrix (R) was computed by estimating the covariance between X and Y. SVD of R was then performed, and by projecting the resulting left and right singular vectors onto the X or Y matrices 12 pairs of LVs (four DTI parameters × three clinical measures) (McIntosh et al. Reference McIntosh, Bookstein, Haxby and Grady1996; Krishnan et al. Reference Krishnan, Williams, McIntosh and Abdi2011) with corresponding singular values and voxel saliences were computed.
Significance of the LVs was assessed using permutation tests. This involved a Monte-Carlo like procedure in which a new data set (permutation sample) was obtained by randomly reordering the rows of X but leaving Y/design matrix unchanged (McIntosh et al. Reference McIntosh, Bookstein, Haxby and Grady1996; McIntosh & Lobaugh, Reference McIntosh and Lobaugh2004). The permutation sample was then recomputed in the PLSC model thereby obtaining new singular values. This procedure was repeated 5000 times and the likelihood that the singular values obtained in the original analysis were due to random chance was determined (McIntosh et al. Reference McIntosh, Bookstein, Haxby and Grady1996; McIntosh & Lobaugh, Reference McIntosh and Lobaugh2004).
Stability and reliability of the voxel saliences was determined using bootstrap resampling (Krishnan et al. Reference Krishnan, Williams, McIntosh and Abdi2011). In this procedure a new data set (bootstrap sample) was obtained by resampling with replacement the rows of both X and Y/design matrix. The procedure was repeated 500 times, and the standard error of the saliences at each voxel was estimated. Bootstrap ratios were obtained by dividing each voxels salience by its standard error. Saliences with bootstrap ratios greater than ±1.96 were considered reliable since bootstrap ratios are roughly analogous to Z scores (a Z score of 1.96 has p < 0.05, two-tailed) (Grady et al. Reference Grady, McIntosh, Horwitz and Rapoport2000; Ziegler et al. Reference Ziegler, Dahnke, Winkler and Gaser2013; Cha et al. Reference Cha, Fekete, Siciliano, Biezonski, Greenhill, Pliszka, Blader, Roy, Leibenluft and Posner2015). We reported reliable bootstrap clusters (cluster size ⩾25 voxels) (Ersche et al. Reference Ersche, Williams, Robbins and Bullmore2013) with positive and negative correlations to the pattern seen in LV. Coordinates in all tables were extracted from JHU WM tractography atlas provided by FSL (Mori & van, Reference Mori and van Zijl2007; Wakana et al. Reference Wakana, Caprihan, Panzenboeck, Fallon, Perry, Gollub, Hua, Zhang, Jiang, Dubey, Blitz, van Zijl and Mori2007; Hua et al. Reference Hua, Zhang, Wakana, Jiang, Li, Reich, Calabresi, Pekar, van Zijl and Mori2008) and reported in Montreal Neurological Institute (MNI) co-ordinates.
Statistical analysis of MRI data using univariate voxelwise analysis and ROI
In order to confirm the direction and localization of the PLSC findings we used univariate voxelwise analysis to compare DTI maps (FA, AD, MO, RD, and MD) between UHR individuals and HCs, and correlated the DTI maps to positive and negative symptoms (CAARMS and SANS) and level of functioning (SOFAS) within the UHR individuals. The univariate analysis were performed non-parametrically using the randomize function (5000 permutations) from the FSL FMRIB software library v5.0.9 (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki) (Smith et al. Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg, Bannister, De Luca, Drobnjak, Flitney, Niazy, Saunders, Vickers, Zhang, De Stefano, Brady and Matthews2004; Woolrich et al. Reference Woolrich, Jbabdi, Patenaude, Chappell, Makni, Behrens, Beckmann, Jenkinson and Smith2009; Jenkinson et al. Reference Jenkinson, Beckmann, Behrens, Woolrich and Smith2012). Family-wise error correction (Bullmore et al. Reference Bullmore, Suckling, Overmeyer, Rabe-Hesketh, Taylor and Brammer1999) with a threshold of p < 0.05 was used to correct for multiple comparisons and threshold-free cluster enhancement (Smith & Nichols, Reference Smith and Nichols2009) was applied to enhance cluster-like structures but keeping the images fundamentally voxelwise. The anatomical locations of significant regions were identified in MNI co-ordinates using the JHU WM tractography atlas (Mori & van, Reference Mori and van Zijl2007; Wakana et al. Reference Wakana, Caprihan, Panzenboeck, Fallon, Perry, Gollub, Hua, Zhang, Jiang, Dubey, Blitz, van Zijl and Mori2007; Hua et al. Reference Hua, Zhang, Wakana, Jiang, Li, Reich, Calabresi, Pekar, van Zijl and Mori2008).
To quantify the results found in the PLSC analysis we used the three largest reliable spatial regions with positive correlations and the largest reliable spatial region with negative correlations to the pattern in LV1 as ROIs. Mean FA, AD, MO, and RD in these ROIs were compared between UHR individuals and HCs.
Results
Two of the 50 individuals included in the study were not MRI scanned due to dental braces, two did not undergo the DTI-sequences due to technical problems, and one had to be excluded from analyses due to poor MRI quality. Thus, 45 UHR individuals were included in the current analyses. Forty-five HCs were included; however, in one HC only one of the two DWI acquisitions could be used.
Sample characteristics
The UHR individuals and HCs did not differ significantly regarding handedness, height, weight, alcohol consumption, or use of benzodiazepines. The groups differed significantly in years of education, parental socioeconomic status, tobacco smoking, and drug abuse. Demographic and clinical data are presented in Table 1.
WM differences between UHR and HCs
Multivariate PLSC analysis of UHR individuals v. HCs identified one significant LV with p < 0.001 accounting for 52% of the cross-block covariance. This LV showed a pattern where the UHR individuals had lower FA, AD, and MO along with higher RD in widespread brain regions (displayed in red-yellow color in Fig. 1). One smaller region showed the inverse pattern with higher FA, AD, and MO along with lower RD (displayed in blue color in Fig. 1). Each DTI parameter contributed significantly to this pattern as demonstrated by the error-bars (based on bootstrapping); see Fig. 1 and Table 2.
Multivariate analysis of white matter differences between UHR and healthy controls. It shows white matter differences between UHR individuals and healthy controls. The panel to the left shows the pattern identified by LV1, in which UHR individuals have lower FA, AD, and MO concomitant with higher RD than healthy controls. In the panel to the right the most reliable spatial regions demonstrating this pattern (i.e. positive bootstrap-ratios) are highlighted in red-yellow color. Spatial regions demonstrating the inverse pattern (i.e. negative bootstrap ratios) are highlighted in blue color. Only clusters with bootstrap ratios greater than ±1.96 and cluster size ⩾25 voxels are shown. The brain is displayed according to radiological convention (participant's left is to the right) and in axial slices. Clusters are enhanced using tbss_fill (FSL version 5.09).
Multivariate analysis of white matter differences between UHR and healthy controls
NA, not applicable
Table 2 shows the most reliable spatial brain regions with white matter differences between UHR individuals and healthy controls. In regions with positive correlations to the pattern in LV1 (i.e. positive bootstrap-ratios) the UHR individuals have: ↓FA, ↓AD, ↓MO, ↑RD compared with healthy controls, and in regions with negative correlations (i.e. negative bootstrap-ratios) to the pattern in LV1 the UHR individuals have: ↑FA, ↑AD, ↑MO, ↓RD compared with healthy controls. Only regions with bootstrap ratios greater than ±1.96 and cluster size ⩾25 voxels are displayed
Localization is performed using FSL
a Local maxima in MNI coordinates (mm) are extracted from JHU WM tractography atlas
Univariate voxelwise analyses of UHR individuals v. HCs found significantly lower FA in the UHR individuals in brain regions involving the superior longitudinal fasciculus, inferior fronto-occipital fasciculus, and anterior thalamic radiation. No significant voxelwise differences in AD, MO, and RD maps between UHR individuals and HCs were found. See online Supplementary Fig. S1.
When quantifying the PLSC results we compared mean FA, AD, MO, and RD in UHR individuals v. HCs in the three largest reliable clusters (used as ROIs) found by PLSC with positive correlations and the largest cluster with negative correlations (only one was identified) to the pattern in LV. This showed significantly different mean values for the majority of DTI parameters. See online Supplementary Table S2.
Associations between DTI parameters and clinical measures
Multivariate PLSC analysis correlating DTI parameters to clinical scores revealed five significant LVs (all permuted p values < 0.03). LV1–LV5 accounted for 31, 18, 15, 8, and 5% of the cross-block covariance, respectively. Since LV1 explained a considerably larger proportion of the cross-block covariance, only LV1 is reported here. The pattern identified by LV1 showed that more positive and negative symptoms (i.e. high CAARMS and SANS scores) and low levels of functioning (i.e. low SOFAS scores) were correlated to lower FA, AD, MO, and higher RD (regions displayed in red-yellow color in Fig. 2). A few brain regions demonstrated the inverse correlations with more positive and negative symptoms (i.e. high CAARMS and SANS scores) combined with low levels of functioning (i.e. low SOFAS scores) being correlated to higher FA, AD, MO, and lower RD (regions displayed in blue color in Fig. 2). The error-bars (based on bootstrapping) in Fig. 2 show that each DTI parameter contributes significantly to the pattern in LV1. See Fig. 2 and Table 3.
Multivariate analysis associating white matter to symptoms and level of functioning within UHR individuals. It shows the associations found by LV1 between DTI parameters, symptoms (CAARMS and SANS), and level of functioning (SOFAS) within the UHR individuals. The panel to the left shows the pattern identified by LV1 in which more positive and negative symptoms (high CAARMS and SANS scores) and low levels of functioning (low SOFAS scores) are correlated to ↓FA, ↓AD, ↓MO and ↑ RD. In the panel to the right the most reliable spatial regions demonstrating this pattern (i.e. positive bootstrap-ratios) are highlighted in red-yellow color. Spatial regions demonstrating the inverse pattern (i.e. negative bootstrap ratios) are highlighted in blue color. Only clusters with bootstrap ratios greater than ±1.96 and cluster size ⩾25 voxels are shown. The brain is displayed according to radiological convention (participant's left is to the right) and in axial slices. Clusters are enhanced using tbss_fill (FSL version 5.09).
Multivariate analysis associating white matter to symptoms and level of functioning within UHR individuals
NA, not applicable
Table 3 shows the most reliable spatial brain regions in which DTI parameters are correlated to symptoms (CAARMS and SANS), and level of functioning (SOFAS) within the UHR individuals. In regions with positive correlations (i.e. positive bootstrap-ratios) to the pattern in LV1, more positive and negative symptoms (i.e. high CAARMS and SANS scores) and low levels of functioning (i.e. low SOFAS scores) are correlated to ↓FA, ↓AD, ↓MO, and ↑RD, while regions with negative correlations (i.e. negative bootstrap-ratios) to LV1 show the inverse pattern. Only brain areas with bootstrap ratios greater than ±1.96 and cluster size⩾25 voxels are displayed
Localization is performed using FSL
a Local maxima in MNI coordinates (mm) are extracted from JHU WM tractography atlas
Univariate voxelwise analysis correlating DTI maps to positive and negative symptoms (CAARMS and SANS) and level of functioning (SOFAS) showed that SOFAS was significantly positively correlated to FA in brain regions involving the left inferior longitudinal fasciculus, that SOFAS was positively correlated to AD in brain regions involving the left inferior fronto-occipital fasciculus and the right superior longitudinal fasciculus, and that SOFAS was positively correlated to MO in brain regions involving the right inferior fronto-occipital fasciculus, the left superior longitudinal fasciculus, and the right anterior thalamic radiation. No significant correlations between SOFAS and RD, and MD were found, and SOFAS was not significantly negatively correlated to any of the DTI parameters. Neither SANS nor CAARMS was significantly correlated (positively or negatively) to any of the DTI parameters. See online Supplementary Fig. S2.
Discussion
The present study is the first to use PLSC to explore WM microstructure in UHR individuals. In the PLSC group analysis one pattern of DTI parameters could significantly differentiate UHR individuals from matched HCs. This pattern indicated that UHR individuals had lower FA, AD, MO, and higher RD in widespread brain regions compared with HCs. Similar findings of low FA concomitant with high RD have been reported in several studies in UHR, in studies of first-episode of psychosis, and in studies of chronic schizophrenia (von Hohenberg et al. Reference von Hohenberg, Pasternak, Kubicki, Ballinger, Vu, Swisher, Green, Giwerc, Dahlben, Goldstein, Woo, Petryshen, Mesholam-Gately, Woodberry, Thermenos, Mulert, McCarley, Seidman and Shenton2014; Spalletta et al. Reference Spalletta, De Rossi, Piras, Iorio, Dacquino, Scanu, Girardi, Caltagirone, Kirkpatrick and Chiapponi2015; Zhu et al. Reference Zhu, Zhuo, Qin, Wang, Ma, Zhou and Yu2015; Li et al. Reference Li, Xie, Liu, Song, Chen, Li, Lu, Lv, Wang, Yan, Yan, Zhang, Zhang and Jiang2016; van Dellen et al. Reference van Dellen, Bohlken, Draaisma, Tewarie, van Lutterveld, Mandl, Stam and Sommer2016; Zeng et al. Reference Zeng, Ardekani, Tang, Zhang, Zhao, Cui, Fan, Zhuo, Li, Xu, Goff and Wang2016). Findings of low AD are more scarce, and have been reported in genetic- and clinical high-risk cohorts (Kikinis et al. Reference Kikinis, Makris, Finn, Bouix, Lucia, Coleman, Tworog-Dube, Kikinis, Kucherlapati, Shenton and Kubicki2013; Rigucci et al. Reference Rigucci, Santi, Corigliano, Imola, Rossi-Espagnet, Mancinelli, De Pisa, Manfredi, Bozzao, Carducci, Girardi and Comparelli2016) and in early psychosis (Hatton et al. Reference Hatton, Lagopoulos, Hermens, Hickie, Scott and Bennett2014).
In DTI, the most frequently used summary measure is FA. However, other DTI parameters such as AD, RD, and MO may reflect more distinct aspects of WM microstructure (Aung et al. Reference Aung, Mar and Benzinger2013). Axonal damage such as axonal swelling (Song et al. Reference Song, Yoshino, Le, Lin, Sun, Cross and Armstrong2005; Kim et al. Reference Kim, Loy, Wang, Budde, Schmidt, Trinkaus and Song2010) or diffuse axonal injury (Arfanakis et al. Reference Arfanakis, Haughton, Carew, Rogers, Dempsey and Meyerand2002; Li et al. Reference Li, Li, Feng and Gu2011; Bennett et al. Reference Bennett, Mac Donald and Brody2012) has been shown to be reflected by a decrease in AD. Demyelination has been associated with an increase in RD (Song et al. Reference Song, Sun, Ramsbottom, Chang, Russell and Cross2002, Reference Song, Yoshino, Le, Lin, Sun, Cross and Armstrong2005; Harsan et al. Reference Harsan, Poulet, Guignard, Steibel, Parizel, de Sousa, Boehm, Grucker and Ghandour2006; Sun et al. Reference Sun, Liang, Le, Armstrong, Cross and Song2006a , Reference Sun, Liang, Trinkaus, Cross, Armstrong and Song b ; Xie et al. Reference Xie, Tobin, Budde, Chen, Trinkaus, Cross, McDaniel, Song and Armstrong2010) sometimes preceded or accompanied by a decrease in AD probably due to minor axonal damage early in the demyelination process (Harsan et al. Reference Harsan, Poulet, Guignard, Steibel, Parizel, de Sousa, Boehm, Grucker and Ghandour2006; Sun et al. Reference Sun, Liang, Trinkaus, Cross, Armstrong and Song2006b ; Tyszka et al. Reference Tyszka, Readhead, Bearer, Pautler and Jacobs2006). We speculate that our findings of high RD accompanied by low AD could reflect abnormalities in the myelination process, which could in turn sub-serve the observed symptomatology. In our study, the abnormalities in FA, AD, and RD were accompanied by abnormalities in MO. This is in agreement with previous studies finding an association between low FA, high RD, and low MO in patients with schizophrenia (Kumar et al. Reference Kumar, Iwabuchi, Oowise, Balain, Palaniyappan and Liddle2015; Wigand et al. Reference Wigand, Kubicki, Clemm von, Leicht, Karch, Eckbo, Pelavin, Hawley, Rujescu, Bouix, Shenton and Mulert2015). Since MO has been found to be low in regions of crossing fibers from two fiber populations and high in regions where only one fiber orientation predominates (Ennis & Kindlmann, Reference Ennis and Kindlmann2006; Kindlmann et al. Reference Kindlmann, Ennis, Whitaker and Westin2007; Wigand et al. Reference Wigand, Kubicki, Clemm von, Leicht, Karch, Eckbo, Pelavin, Hawley, Rujescu, Bouix, Shenton and Mulert2015) our findings could be indicative of aberrations in the underlying fiber structure in UHR individuals.
In the PLSC analysis associating DTI parameters to symptoms within the UHR individuals, the pattern explaining the largest proportion of the cross-block covariance (LV1) indicated that severity of symptoms was negatively correlated to FA, AD, MO, and positively to RD in the majority of identified brain regions. The negative correlation between positive symptoms and FA is in accordance with findings in other studies of UHR individuals (Bloemen et al. Reference Bloemen, de Koning, Schmitz, Nieman, Becker, de Haan, Dingemans, Linszen and van Amelsvoort2010), individuals experiencing sub-threshold psychotic symptoms (Cooper et al. Reference Cooper, Alm, Olson and Ellman2016), and patients with first-episode psychosis and chronic schizophrenia (Skelly et al. Reference Skelly, Calhoun, Meda, Kim, Mathalon and Pearlson2008; Catani et al. Reference Catani, Craig, Forkel, Kanaan, Picchioni, Toulopoulou, Shergill, Williams, Murphy and McGuire2011; Cui et al. Reference Cui, Chen, Deng, Huang, Li, Ma, Huang, Jiang, Wang, Wang, Collier, Gong and Li2011; Boos et al. Reference Boos, Mandl, van Haren, Cahn, van Baal, Kahn and Hulshoff Pol2013; Hatton et al. Reference Hatton, Lagopoulos, Hermens, Hickie, Scott and Bennett2014; Ohtani et al. Reference Ohtani, Bouix, Lyall, Hosokawa, Saito, Melonakos, Westin, Seidman, Goldstein, Mesholam-Gately, Petryshen, Wojcik and Kubicki2015; Ebdrup et al. Reference Ebdrup, Raghava, Nielsen, Rostrup and Glenthoj2016; Seitz et al. Reference Seitz, Zuo, Lyall, Makris, Kikinis, Bouix, Pasternak, Fredman, Duskin, Goldstein, Petryshen, Mesholam-Gately, Wojcik, McCarley, Seidman, Shenton, Koerte and Kubicki2016; Zhang et al. Reference Zhang, Fan, Chen, Tan, Tan, Hu, Salas, Kosten, Zunta-Soares and Soares2016). Of note, these studies find a negative correlation between positive symptoms and FA in some of the same areas where we also find a negative correlation namely the left and right superior longitudinal fasciculus (Skelly et al. Reference Skelly, Calhoun, Meda, Kim, Mathalon and Pearlson2008; Ebdrup et al. Reference Ebdrup, Raghava, Nielsen, Rostrup and Glenthoj2016) and the right inferior longitudinal fasciculus (Seitz et al. Reference Seitz, Zuo, Lyall, Makris, Kikinis, Bouix, Pasternak, Fredman, Duskin, Goldstein, Petryshen, Mesholam-Gately, Wojcik, McCarley, Seidman, Shenton, Koerte and Kubicki2016). Most of the studies only report on FA, but a few studies also report a positive correlation between positive symptoms and RD and a negative correlation between positive symptoms and MO (Wigand et al. Reference Wigand, Kubicki, Clemm von, Leicht, Karch, Eckbo, Pelavin, Hawley, Rujescu, Bouix, Shenton and Mulert2015; Seitz et al. Reference Seitz, Zuo, Lyall, Makris, Kikinis, Bouix, Pasternak, Fredman, Duskin, Goldstein, Petryshen, Mesholam-Gately, Wojcik, McCarley, Seidman, Shenton, Koerte and Kubicki2016).
Our findings of a negative correlation between severity of negative symptoms and FA are consistent with many studies in first-episode of psychosis and schizophrenia (Wolkin et al. Reference Wolkin, Choi, Szilagyi, Sanfilipo, Rotrosen and Lim2003; Szeszko et al. Reference Szeszko, Robinson, Ashtari, Vogel, Betensky, Sevy, Ardekani, Lencz, Malhotra, McCormack, Miller, Lim, Gunduz-Bruce, Kane and Bilder2008; Asami et al. Reference Asami, Hyuk, Bouix, Rathi, Whitford, Niznikiewicz, Nestor, McCarley, Shenton and Kubicki2014; Ohtani et al. Reference Ohtani, Bouix, Lyall, Hosokawa, Saito, Melonakos, Westin, Seidman, Goldstein, Mesholam-Gately, Petryshen, Wojcik and Kubicki2015; Sun et al. Reference Sun, Lui, Yao, Deng, Xiao, Zhang, Huang, Hu, Bi, Li, Sweeney and Gong2015; Zeng et al. Reference Zeng, Ardekani, Tang, Zhang, Zhao, Cui, Fan, Zhuo, Li, Xu, Goff and Wang2016). Collectively, our data add to the notion that in UHR individuals, abnormal WM microstructure in widespread brain regions may underlie both positive and negative symptoms and low levels of functioning.
Study limitations
Although our sample is larger than those in most previous DTI studies in UHR individuals, statistical power may still be limited. We addressed this by using PLSC, which limits the need for correction for multiple comparisons (McLelland et al. Reference McLelland, Chan, Ferber and Barense2014; Protzner et al. Reference Protzner, Hargreaves, Campbell, Myers-Stewart, van Sophia, Goodyear, Sargious and Pexman2016). By means of this non-biased approach, we identified clinical meaningful significant group differences as well as with-in group associations. The identified regional overlaps with previous studies of patients with schizophrenia at different stages of illness, add to the external validity of our data-driven, whole brain findings. The UHR individuals and HCs were matched on gender and age (±3 years), however, as could be expected the groups differed in terms of years of education, parental socioeconomic status, tobacco smoking, and drug use. Of these, especially tobacco smoking and drug use could represent potential confounding factors. Effects of tobacco smoking on WM have shown equivocal results. Some studies have reported high FA (Jacobsen et al. Reference Jacobsen, Picciotto, Heath, Frost, Tsou, Dwan, Jackowski, Constable and Mencl2007; Paul et al. Reference Paul, Grieve, Niaura, David, Laidlaw, Cohen, Sweet, Taylor, Clark, Pogun and Gordon2008; Liao et al. Reference Liao, Tang, Deng, Deng, Luo, Wang, Chen, Liu, Chen, Brody and Hao2011; Hudkins et al. Reference Hudkins, O'Neill, Tobias, Bartzokis and London2012; Huang et al. Reference Huang, DiFranza, Kennedy, Zhang, Ziedonis, Ursprung and King2013; van Ewijk et al. Reference van Ewijk, Groenman, Zwiers, Heslenfeld, Faraone, Hartman, Luman, Greven, Hoekstra, Franke, Buitelaar and Oosterlaan2015; Viswanath et al. Reference Viswanath, Velasquez, Thompson-Lake, Savjani, Carter, Eagleman, Baldwin, De La Garza and Salas2015; Yu et al. Reference Yu, Yuan, Zhang, Liu, Dong, Jin, Luo, Zhai, Zhao, Zhao, Gu, Xue, Liu, Lu, Qin and Tian2016), but tobacco use has been associated with low FA in other studies (Lin et al. Reference Lin, Wu, Zhu and Lei2013; Savjani et al. Reference Savjani, Velasquez, Thompson-Lake, Baldwin, Eagleman, De La Garza and Salas2014; Umene-Nakano et al. Reference Umene-Nakano, Yoshimura, Kakeda, Watanabe, Hayashi, Nishimura, Takahashi, Moriya, Ide, Ueda, Hori, Ikenouchi-Sugita, Katsuki, Atake, Abe, Korogi and Nakamura2014; Baeza-Loya et al. Reference Baeza-Loya, Velasquez, Molfese, Viswanath, Curtis, Thompson-Lake, Baldwin, Ellmore, De La Garza and Salas2016). Moreover, it has been suggested that the effect of nicotine may differ across the lifespan with FA increase in young (age < 30 years) and FA decrease in older smokers (Gogliettino et al. Reference Gogliettino, Potenza and Yip2016). Long-term use of drugs such as cannabis (Ashtari et al. Reference Ashtari, Cervellione, Cottone, Ardekani, Sevy and Kumra2009; Bava et al. Reference Bava, Frank, McQueeny, Schweinsburg, Schweinsburg and Tapert2009; Yucel et al. Reference Yucel, Zalesky, Takagi, Bora, Fornito, Ditchfield, Egan, Pantelis and Lubman2010; James et al. Reference James, Hough, James, Winmill, Burge, Nijhawan, Matthews and Zarei2011; Zalesky et al. Reference Zalesky, Solowij, Yucel, Lubman, Takagi, Harding, Lorenzetti, Wang, Searle, Pantelis and Seal2012), inhalants (Yucel et al. Reference Yucel, Zalesky, Takagi, Bora, Fornito, Ditchfield, Egan, Pantelis and Lubman2010), and cocaine (Lim et al. Reference Lim, Choi, Pomara, Wolkin and Rotrosen2002; Moeller et al. Reference Moeller, Hasan, Steinberg, Kramer, Dougherty, Santos, Valdes, Swann, Barratt and Narayana2005; Ma et al. Reference Ma, Hasan, Steinberg, Narayana, Lane, Zuniga, Kramer and Moeller2009; Lane et al. Reference Lane, Steinberg, Ma, Hasan, Kramer, Zuniga, Narayana and Moeller2010; Narayana et al. Reference Narayana, Herrera, Bockhorst, Esparza-Coss, Xia, Steinberg and Moeller2014) has primarily been associated to low FA and high RD. Post hoc analyses comparing the subgroup of 26 UHR individuals, who never had an abuse or dependency of alcohol, drugs, or tobacco to the HCs revealed a significant (p < 0.001) pattern similar to the one seen in the analysis including all 45 UHR individuals, suggesting that these confounders have not substantially influenced the present results. See online Supplementary Fig. S3.
Some of the UHR individuals had received antipsychotic medication, although in very modest cumulative lifetime doses (not exceeding 50 mg haloperidol or equivalent other antipsychotic medication). Post hoc analyses comparing the subgroup of 36 UHR individuals, who had never taken antipsychotic medication to the HCs revealed a significant (p < 0.001) pattern similar to the one found in the analysis, including all 45 UHR individuals. See online Supplementary Fig. S4. Post hoc analyses comparing the subgroup of 33 UHR individuals who did not take antidepressant medication to the HCs showed a significant pattern (p < 0.001) similar to the one including all 45 UHR individuals. See online Supplementary Fig. S5. These findings suggest that exposure to neither antipsychotic nor antidepressant medication substantially influenced the present results.
Interpretation of DTI findings can be challenging. The measured diffusion effects are averaged over a voxel with much larger dimensions (measured in mm) than individual axons (measured in μm). Therefore, it can be difficult to disentangle signals reflecting aberrations in myelination from signals reflecting aberrations in fiber architecture, since both will influence RD and MO. Future MRI studies in UHR may benefit from combining DTI with other imaging contrasts, such as magnetization transfer ratio or myelin water fraction, which are putative measures of WM myelination (Mandl et al. Reference Mandl, Pasternak, Cahn, Kubicki, Kahn, Shenton and Hulshoff Pol2015).
When visualizing the most reliable spatial regions showing the identified WM patterns we designated a bootstrap ratio threshold of ±1.96 (corresponding to approximately p < 0.05). However, since the PLSC uses the entire set of voxels as a single multivariate representation the exact extent and localization of the identified WM patterns are difficult to ascertain. We addressed this by doing additional univariate voxelwise analysis and by quantifying the mean values of the DTI parameters in the largest spatial regions found by the PLSC (used as ROIs).
Since the contrast in DTI is generated by measuring the microscopic diffusion of water molecules in brain tissue it is sensitive to macroscopic head motion (Ling et al. Reference Ling, Merideth, Caprihan, Pena, Teshiba and Mayer2012; Yendiki et al. Reference Yendiki, Koldewyn, Kakunoori, Kanwisher and Fischl2014). It seems that more head motion is associated to an overestimation of RD and underestimation of AD and FA (Yendiki et al. Reference Yendiki, Koldewyn, Kakunoori, Kanwisher and Fischl2014). In the present study foam-pads inside the scanner were used to minimize head motion, nevertheless more absolute but not relative head motion was observed in the UHR individuals compared with the HCs. No linear correlation was observed between relative head motion and CAARMS, SANS, or SOFAS [Pearson correlation = 0.006/0.049/−0.234 and p = 0.968/0.749/0.122 (2-sided)] or between absolute head motion and CAARMS or SANS [Pearson correlation = 0.167/0.110 and p = 0.272/0.941 (2-sided)]. However there was a significant correlation between absolute head motion and SOFAS [Pearson correlation = −0.298 and p = 0.047 (2-sided)]. We addressed this by visually inspecting all DW images before processing and discharging those where head motion led to visibly poor quality. Additionally DW images were corrected for eddy current and head-motion as part of the FSL processing pipeline, and furthermore we corrected for head motion parameters (absolute and relative) in both the univariate and the multivariate analysis.
Conclusion
In conclusion, we confirm the presence of subtle abnormalities in WM microstructure in UHR individuals with minimal antipsychotic exposure. The UHR individuals displayed a pattern of low FA, AD, and MO, concomitant with high RD compared with HCs. Within UHR individuals, the same pattern was associated to both positive and negative symptoms as well as level of functioning. Cerebral dysmyelination or aberrations in WM fiber architecture may play an important role in the emerging symptoms constituting a vulnerability for developing a psychotic disorder.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291717001210.
Acknowledgements
Dr Ebdrup has received lecture fees and/or is part of Advisory Boards of Bristol-Myers Squibb, Eli Lilly and Company, Janssen-Cilag, Otsuka Pharma Scandinavia and Takeda Pharmaceutical Company. Dr Glenthøj is the leader of a Lundbeck Foundation Center of Excellence for Clinical Intervention and Neuropsychiatric Schizophrenia Research (CINS), which is partially financed by an independent grant from the Lundbeck Foundation based on international review and partially financed by the Mental Health Services in the Capital Region of Denmark, the University of Copenhagen, and other foundations. All grants are the property of the Mental Health Services in the Capital Region of Denmark and administrated by them. This work was supported by the Danish Council for Independent Research, Medical Sciences (DFF-4004-00314), the Lundbeck Foundation (R25-A2701 and R155-2013-16337), the University of Copenhagen (BHE, no grant number), and Mental Health Center Copenhagen (KK, no grant number).
Declaration of Interest
None.
Ethical Standard
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional guides on the care and use of laboratory animals.
