In our study, we analyzed the agreement between CSF biomarkers (Aβ1-42, pTau, tTau) and their ratios (Aβ1-42/Aβ1-40, Aβ1-42/pTau, Aβ1-42/tTau,) with amyloid-PET in a heterogeneous sample of participants from 5 centers. Different laboratory protocols, CSF analysis techniques (Innotest®, Lumipulse®), biomarker cut-off points and amyloid-PET tracers (18F-Florbetaben, 18F-Flutemetamol) were used. Regarding the diagnoses, there were participants with neurodegenerative diseases, non-neurodegenerative diseases, subjective memory complaints and cognitively unimpaired individuals. Both a LP and an amyloid-PET scan had been performed, in the framework of clinical trials, research projects or a non-conclusive first test. In the five centers, the use of a second CSF biomarker resulted in an increase of the agreement with amyloid-PET. The highest agreement was found in participants with an A + T + N + profile.

Clinical indication of AD CSF biomarkers and amyloid-PET is similar and there are no studies that show if one test is preferable over the other one in clinical practice. It’s well known that lumbar puncture is more invasive, although well tolerated [25], and that amyloid-PET is more expensive. A recent study has explored cost-effectiveness of these two tests in the diagnosis of AD among subjects with early onset cognitive decline, concluding that amyloid-PET is not a cost-effective technique compared to AD CSF biomarkers [26]. Both share the need to establish thresholds for positivity, leading to a binary separation of positive and negative results, while the AD pathophysiological process is more complex, and each biomarker has its own and different trajectory [27]. The LP has the advantage of providing information about the three components of the ATN framework in one test. Other advantages of obtaining CSF through a LP are to allow analyzing other biomarkers such as α-synuclein, 14–3-3 protein or neurofilament light chain, which could be especially useful to complete the diagnostic study if AD CSF biomarkers are negative.

Aβ1-42/Aβ1-40, pTau/Aβ1-42 and tTau/Aβ1-42 ratios showed a better agreement with amyloid-PET than individual biomarkers. These results are in line with recent studies [5, 28,29,30] demonstrating that the use of CSF ratios improves agreement with amyloid-PET over using single biomarkers. Firstly, normalizing Aβ1-42 to the concentration of Aβ1-40, a peptide that is much more abundant in CSF, may compensate individual differences in amyloid precursor protein processing and provide a more specific information on the pathological amyloidosis deposition [31]. In this line, a recent work demonstrated that the Aβ1–42/Aβ1–40 ratio in CSF is more strongly associated to tau markers and clinical progression than Aβ1–42 alone [32]. In the same way, another recent work showed that global cortex standardized uptake value ratios of amyloid deposition in amyloid-PET correlated highly with CSF Aβ1-42/Aβ1-40 and moderately with Aβ1-42 but not with Aβ1-40 [30]. Our results replicate those of Amft et al., that made a comparison between amyloid-PET and CSF biomarkers with pre-defined cut-offs in a clinical cohort with memory deficits, showing that combined biomarkers in CSF, specially pTau/Aβ1-42 and Aβ1-42/Aβ1-40, predicted amyloid-PET result better than Aβ1-42 [33]. Secondly, it has been described that CSF Aβ1-42 levels can be abnormal earlier in the disease course [34] than amyloid-PET visual read. Therefore, combining Aβ1-42 in a ratio with pTau or tTau, markers that are abnormal later in the disease, may correspond better to amyloid-PET visual read.

When comparing ATN profiles, in our study participants with an A + T + N + or A-T-N-profile had the highest agreement between Aβ1-42 and amyloid-PET. About 2/3 of the participants with a A-T + N + profile had a positive amyloid-PET, and 50% of the participants with A + T-N- profile had a positive amyloid-PET. In the last group, the positivity of Aβ1-42 in CSF but not in the amyloid-PET could be explained by a low concentration of total amyloid peptides (that corrects by normalizing to the concentration of Aβ1-40) or a very initial phase of AD, in which Aβ1-42 becomes positive in CSF before it does in amyloid-PET.

Several reasons may contribute to the limitation of agreement between both methods in this study: as we mentioned, the clearcut separation between negative and positive patients in relation to a given biomarker is somehow artificial and differs between sites and studies. In our study, when we eliminated the borderline Aβ1-42 and Aβ1-42/Aβ1-40 values from the analysis, we found small changes in the agreement between CSF amyloid biomarkers and amyloid-PET. Therefore, the lack of agreement between the two techniques could not be attributed to the borderline values or could explain only a small part of this lack of agreement. Regarding CSF analysis techniques, automated platforms such as Lumipulse® reduce manual steps as a source of variation; recent studies directly comparing measurements with Lumipulse® with Innotest® showed reduced intra- and inter-assay variability on the Lumipulse® [35]. However, in this study, 34% of CSF biomarkers were determined using Innotest®. In our study, measurements of Aβ1-42 and pTau made with Lumipulse ® showed a better agreement with amyloid-PET than those made with Innotest®. However, this is not a study designed to compare these two techniques and there are confounding factors (such as the reason for which the two tests were performed) that are not equally distributed in the two groups. Finally, regarding amyloid-PET analysis, the use of the centiloid quantification scale instead of the visual read may reduce the variability and identify earlier stages of amyloid accumulation [36, 37].

In this multicenter study, each of the participating centers developed their own cut-off points for each biomarker maximizing sensitivity and specificity. Therefore, mean values of biomarkers are different and not comparable between centers.

Time between LP and PET-scan can influence the results of the study. As it is well known, CSF AB1-42 starts to decrease in CSF before amyloid accumulation is detected by PET imaging, and both precede CSF p-Tau and t-Tau increase in CSF. Previous studies described that CSF Aβ42 was fully abnormal 5–10 years or more before dementia diagnosis, there was little change in the anatomical extent of amyloid PET over time in individuals with mild AD while it was static by the time the patients became demented. By contrast, both CSF t-tau and p-tau became progressively more abnormal as the time to diagnosis of dementia decreased, in periods of 2.5 years previous to dementia [38, 39]. [Jack, Forster] We set the time limit at 18 months to control that the time difference was not responsible for the lack of agreement between biomarkers. The mean time between both tests was 5 months, a period that would normally elapse in clinical practice between the two tests, and in which we expect no changes or minimal changes in CSF biomarker levels and no changes in the result of the amyloid PET.

This study has some limitations that need to be taken into consideration. In the first place, suspected diagnoses before CSF biomarkers and amyloid PET results were not registered. Therefore, valuable information regarding the agreement of both tests according to initial clinical orientation is lost. Secondly, participants came from different centers attending diverse populations, using different biomarker cut-offs, CSF platform analysis and PET tracers and therefore resembling those of daily clinical practice. But also, some patients participating in clinical assays and research projects may have been carefully selected, with none or few comorbidities, and those with a non-conclusive first tests were probably of high complexity, compared to the characteristics of patients in real-world daily clinical practice. Finally, the number of patients in some of the centers and in some of the diagnostic categories was small.

The main strength of this study is that it is a multicenter study with the participation of 5 centers that used different protocols for CSF biomarker analyses, different cutoffs for biomarkers positivity and different PET tracers, so the results resemble the heterogeneity of daily clinical practice. We are aware of some limitations: diagnoses reported in the study considered the result of biomarkers; since this was a retrospective study, it was not possible to systematically record the clinical diagnosis prior to the biomarkers use. Also, sample size of the 5 centers was different and some of the clinical categories included a small number of participants.

In conclusion, in this heterogeneous multicenter study, combined biomarkers in CSF (Aβ1-42/Aβ1-40, pTau/Aβ1-42, tTau/Aβ1-42) were better markers of cerebral amyloid deposition, as identified by amyloid tracers, than single biomarkers (Aβ1-42, pTau, tTau). Participants with an A + T + N + profile had a high percentage of positive amyloid-PET scans (96%) and those with an A-T-N- profile of negative amyloid-PET scans (89%), whereas participants with an A + T-N- profile had the same proportion of positive (49%) and negative (51%) amyloid-PET scans.

In conclusion, in this heterogeneous multicenter study, combined biomarkers in CSF (Aβ1-42/Aβ1-40, ptau/Aβ1-42, tTau/Aβ1-42) were better markers of cerebral amyloid deposition, as identified by amyloid tracers, than single biomarkers (Aβ1-42, pTau, tTau). In participants with A + T + N + and A-T-N- profiles, CSF Aβ1-42 and amyloid-PET were of the same sign, whereas those with a A + T-N- profile had the same proportion of positive and negative amyloid-PET scans.